initial commit

2026-05-30 23:29:44 -07:00
commit 431ffdff06
48 changed files with 4221 additions and 0 deletions
@@ -0,0 +1,13 @@
+.env
+.venv
+__pycache__
+.pytest_cache
+.ruff_cache
+.mypy_cache
+*.pyc
+storage
+tmp
+dist
+build
+frontend/node_modules
+frontend/dist
@@ -0,0 +1,16 @@
+.env
+config.toml
+.venv/
+__pycache__/
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+*.pyc
+*.pyo
+*.pyd
+*.egg-info/
+dist/
+build/
+storage/
+tmp/
+.DS_Store
@@ -0,0 +1,17 @@
+FROM python:3.12-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends ffmpeg curl fonts-noto-cjk \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY pyproject.toml README.md ./
+COPY src ./src
+COPY alembic.ini ./alembic.ini
+COPY migrations ./migrations
+RUN pip install --no-cache-dir .
+
+CMD ["uvicorn", "evanescere.api:app", "--host", "0.0.0.0", "--port", "8000"]
@@ -0,0 +1,374 @@
+# Evanescere
+
+Evanescere turns recorded livestreams into suggested, rendered clips:
+
+1. Poll IIS WebDAV for finished recordings.
+2. Wait until file size is unchanged across polling cycles.
+3. Download the stable source file to Framework-local storage.
+4. Remux FLV/H264 to MP4 and extract ASR-ready audio.
+5. Transcribe Mandarin audio through a FunASR-compatible API.
+6. Ask DeepSeek for ranked timeline-aware clip suggestions.
+7. Generate subtitles, render clips, generate thumbnails with optional local image generation and VTuber overlay, and optionally upload.
+
+The Framework Arch server is the intended production host. The Hyper-V Arch VM can be used for development, API testing, and database/control-plane work.
+
+## Architecture
+
+Evanescere is split into a static frontend and an API/worker backend.
+
+- `frontend`: React/Vite UI. It builds to static files and is served by nginx in Docker.
+- `src/evanescere/api.py`: FastAPI backend. It serves JSON only.
+- `src/evanescere/scheduler.py`: polling loop for IIS WebDAV.
+- `src/evanescere/jobs.py`: Dramatiq queue actors backed by Redis.
+- `src/evanescere/pipeline.py`: orchestration for media, ASR, LLM, render, thumbnail, and upload stages.
+- `src/evanescere/services`: adapters for WebDAV, ffmpeg, FunASR, DeepSeek, subtitles, thumbnails, artifacts, and upload.
+- `migrations`: Alembic migrations for PostgreSQL.
+
+Runtime data flow:
+
+```text
+IIS WebDAV recordings
+  -> scheduler polls file size
+  -> PostgreSQL records video state
+  -> Redis queues pipeline job
+  -> worker downloads source to configured storage.local_root
+  -> ffmpeg remuxes/extracts audio
+  -> FunASR creates timestamped transcript
+  -> DeepSeek creates clip suggestions
+  -> ffmpeg renders clips/subtitles/thumbnails
+  -> thumbnail service composes frame/generated image + VTuber overlay + title
+  -> uploader adapter runs, or noop for testing
+```
+
+The backend tracks durable state in PostgreSQL. Large media artifacts stay on the Framework SSD under `[storage].local_root`; WebDAV is used for source ingestion and can later be used for pushing final artifacts back to the Windows server.
+
+## Configuration
+
+Runtime configuration is TOML-based. Copy the commented example and edit your private config:
+
+```bash
+cp config.example.toml config.toml
+```
+
+The application looks for config in this order:
+
+1. `/etc/evanescere/config.toml`
+2. `./config.toml`
+3. Built-in development defaults, only if no config file exists.
+
+For Docker/Compose, mount your config file to:
+
+```text
+/etc/evanescere/config.toml
+```
+
+The checked-in `docker-compose.yml` mounts local `./config.toml` there for the API, workers, and scheduler.
+
+## Services And Ports
+
+- Frontend: `http://localhost:3000`
+- Backend API: `http://localhost:8000`
+- API docs: `http://localhost:8000/docs`
+- Redis: `localhost:6379` when exposed by Compose
+- PostgreSQL: only started by Compose when using the `local-db` profile
+
+## Test Environment
+
+Create and edit `config.toml`:
+
+```bash
+cp config.example.toml config.toml
+```
+
+For a self-contained local test stack, use the Compose PostgreSQL profile:
+
+```bash
+docker compose --profile local-db up --build
+```
+
+Run database migrations:
+
+```bash
+docker compose run --rm api alembic upgrade head
+```
+
+Bootstrap existing WebDAV files so old recordings are marked `existing_done` instead of auto-processed:
+
+```bash
+docker compose run --rm api evanescere bootstrap-existing
+```
+
+Useful test commands:
+
+```bash
+docker compose run --rm api evanescere scan
+docker compose run --rm api evanescere videos
+docker compose run --rm api evanescere run-video VIDEO_ID
+docker compose logs -f api scheduler worker-media worker-ai
+```
+
+Automated tests are written with `pytest`. After installing Python dev dependencies:
+
+```bash
+pytest -q
+```
+
+Without installing dependencies locally, the available lightweight checks are:
+
+```bash
+python -m compileall src tests migrations
+docker compose config --quiet
+```
+
+For safe test/debug runs, set these in `config.toml`:
+
+```toml
+[app]
+log_level = "DEBUG"
+
+[defaults]
+upload_enabled = false
+
+[upload]
+adapter = "noop"
+```
+
+`adapter = "noop"` prevents accidental real uploads. `upload_enabled = false` lets you inspect rendered artifacts before adding an uploader.
+
+## Frontend Development
+
+The frontend reads the backend URL from `frontend/public/config.js` at runtime:
+
+```js
+window.__EVANESCERE_FRONTEND_CONFIG__ = {
+  apiBaseUrl: "http://localhost:8000"
+};
+```
+
+Manual frontend development requires npm packages. Install only after you approve package installation:
+
+```bash
+cd frontend
+npm install
+npm run dev
+```
+
+The Vite dev server runs on `http://localhost:5173`. Make sure `[app].cors_origins` includes that URL.
+
+Static build:
+
+```bash
+cd frontend
+npm run build
+```
+
+The production Compose frontend service builds the static app and serves `dist/` through nginx.
+
+## Thumbnail Generation
+
+Each rendered clip can produce three thumbnail artifacts:
+
+- `thumbnail_base`: a representative frame extracted from the source MP4 near the middle of the clip.
+- `thumbnail_generated`: optional output from a local image generation command.
+- `thumbnail_final`: the final composed JPG with optional title and VTuber character overlay.
+
+Default provider:
+
+```toml
+[thumbnail]
+provider = "frame_overlay"
+```
+
+This uses the extracted frame as the background, then programmatically overlays the title and character PNG using Pillow.
+
+To overlay the VTuber character, use a transparent PNG:
+
+```toml
+[thumbnail]
+character_overlay_path = "/data/evanescere/assets/vtuber.png"
+character_scale = 0.42
+character_position = "bottom-right"
+```
+
+Supported positions are `bottom-right`, `bottom-left`, `center-right`, and `center-left`.
+
+For a local image generation model, configure a command provider:
+
+```toml
+[thumbnail]
+provider = "command"
+command = "/opt/evanescere-thumbnail/generate-thumbnail"
+```
+
+The command receives JSON on stdin and must create the file named by `output_path`. This makes it easy to wrap ComfyUI, Stable Diffusion, Flux, or any other local generator. Payload shape:
+
+```json
+{
+  "video_id": 123,
+  "clip_id": 456,
+  "source_frame": "/data/evanescere/videos/123/clips/456/thumbnail_base.jpg",
+  "output_path": "/data/evanescere/videos/123/clips/456/thumbnail_generated.png",
+  "width": 1920,
+  "height": 1080,
+  "title_zh": "标题",
+  "summary_zh": "摘要",
+  "reason": "推荐原因",
+  "tags": ["tag"],
+  "transcript": [
+    {"start_sec": 10.0, "end_sec": 15.0, "text": "片段台词"}
+  ]
+}
+```
+
+After the command writes `thumbnail_generated.png`, Evanescere still applies the same title and VTuber overlay to produce `thumbnail_final.jpg`.
+
+## Production Deployment
+
+On the Framework Arch server:
+
+1. Install Docker/Compose.
+2. Copy `config.example.toml` to `config.toml`.
+3. Set `[database].url` to the reachable PostgreSQL server.
+4. Set `[redis].url = "redis://redis:6379/0"` if using Compose Redis.
+5. Set `[storage].local_root = "/data/evanescere"` and mount a Framework SSD directory there.
+6. Set `[webdav]` credentials and `base_url`.
+7. Set `[funasr].base_url` to the local FunASR service.
+8. Set `[deepseek].api_key`.
+9. Keep `[upload].adapter = "noop"` until the full pipeline has been tested.
+
+Start production services:
+
+```bash
+docker compose up -d --build
+docker compose run --rm api alembic upgrade head
+docker compose run --rm api evanescere bootstrap-existing
+```
+
+Then watch logs:
+
+```bash
+docker compose logs -f api scheduler worker-media worker-ai
+```
+
+If you use your existing PostgreSQL server instead of the Compose `local-db` profile, do not start the profile. Ensure PostgreSQL accepts connections from the Framework host and that `[database].url` uses the reachable host/IP.
+
+## Config Reference
+
+The full commented example lives in `config.example.toml`. These are the keys the backend reads:
+
+| Section | Key | Default | Meaning |
+| --- | --- | --- | --- |
+| `app` | `env` | `dev` | Free-form environment label included in logs. Use `prod` on production. |
+| `app` | `log_level` | `INFO` | Python logging level. Use `DEBUG` for pipeline bring-up. |
+| `app` | `log_sql` | `false` | Enables SQLAlchemy SQL logging and engine echo. Noisy in production. |
+| `app` | `cors_origins` | local frontend URLs | Array of frontend origins allowed to call FastAPI. |
+| `database` | `url` | local PostgreSQL URL | SQLAlchemy PostgreSQL connection string. |
+| `redis` | `url` | local Redis URL | Redis broker URL for Dramatiq jobs. |
+| `storage` | `local_root` | `./storage` | Root directory for downloaded sources, remuxed MP4s, audio, transcripts, subtitles, clips, and thumbnails. |
+| `webdav` | `base_url` | local placeholder | IIS WebDAV directory containing recording files. |
+| `webdav` | `username` | empty | WebDAV basic-auth username, if required. |
+| `webdav` | `password` | empty | WebDAV basic-auth password, if required. |
+| `webdav` | `verify_tls` | `true` | Whether WebDAV HTTPS certificates are verified. |
+| `webdav` | `poll_interval_seconds` | `60` | Scheduler interval. A file is stable after two equal positive size samples. |
+| `funasr` | `base_url` | local FunASR URL | FunASR-compatible API base URL. The client calls `/audio/transcriptions`. |
+| `funasr` | `api_key` | empty | Optional bearer token for the FunASR service. |
+| `funasr` | `model` | `paraformer-zh` | Model name sent to FunASR. |
+| `deepseek` | `base_url` | `https://api.deepseek.com` | OpenAI-compatible DeepSeek API base URL. |
+| `deepseek` | `api_key` | empty | DeepSeek API key. Required when clip suggestion is enabled. |
+| `deepseek` | `model` | `deepseek-v4-pro` | Model used for clip suggestion. |
+| `deepseek` | `temperature` | `0.2` | Sampling temperature for clip suggestion. |
+| `defaults` | `suggest_enabled` | `true` | Initial setting for automatic LLM clip suggestion. |
+| `defaults` | `render_enabled` | `true` | Initial setting for automatic rendering. |
+| `defaults` | `upload_enabled` | `true` | Initial setting for automatic upload after render. For testing, set false. |
+| `defaults` | `preserve_final_artifacts` | `true` | Whether rendered final artifacts are marked for preservation. |
+| `defaults` | `bake_subtitles` | `true` | Whether rendered clips include hard-baked subtitles by default. |
+| `clip` | `min_seconds` | `30` | Minimum LLM clip duration accepted by backend. |
+| `clip` | `max_seconds` | `360` | Maximum LLM clip duration accepted by backend. |
+| `clip` | `transcript_chunk_seconds` | `900` | Transcript seconds sent to DeepSeek per request. |
+| `thumbnail` | `enabled` | `true` | Enables thumbnail generation during clip render. |
+| `thumbnail` | `provider` | `frame_overlay` | `frame_overlay` or `command`. |
+| `thumbnail` | `width` | `1920` | Final thumbnail width in pixels. |
+| `thumbnail` | `height` | `1080` | Final thumbnail height in pixels. |
+| `thumbnail` | `character_overlay_path` | empty | Optional transparent PNG of the VTuber character. |
+| `thumbnail` | `character_scale` | `0.42` | Character overlay height as a fraction of final thumbnail height. |
+| `thumbnail` | `character_position` | `bottom-right` | Character placement. |
+| `thumbnail` | `title_enabled` | `true` | Draws the clip title onto the final thumbnail. |
+| `thumbnail` | `command` | empty | Local image generation command used when `provider = "command"`. |
+| `upload` | `adapter` | `noop` | `noop` or `command`. |
+| `upload` | `command` | empty | Command invoked when `adapter = "command"`; receives JSON metadata on stdin. |
+
+Frontend runtime config:
+
+| File | Key | Meaning |
+| --- | --- | --- |
+| `frontend/public/config.js` | `apiBaseUrl` | URL the static frontend uses to call the backend. Change this when hosting frontend/backend on different hosts. |
+
+## CLI
+
+```bash
+evanescere scan
+evanescere bootstrap-existing
+evanescere videos
+evanescere run-video VIDEO_ID
+evanescere transcribe VIDEO_ID
+evanescere suggest VIDEO_ID
+evanescere render CLIP_ID
+evanescere upload CLIP_ID
+```
+
+Inside Compose, prefix commands with:
+
+```bash
+docker compose run --rm api
+```
+
+Useful API endpoints for generated files:
+
+- `GET /videos/{video_id}/artifacts`
+- `GET /clips/{clip_id}/artifacts`
+- `GET /artifacts/{artifact_id}`
+
+Thumbnail outputs are visible through the artifact endpoints. Upload command payloads include `thumbnail` when a `thumbnail_final` artifact exists.
+
+## Debugging
+
+Set verbose logs in `config.toml`:
+
+```toml
+[app]
+log_level = "DEBUG"
+log_sql = false
+```
+
+Then inspect service logs:
+
+```bash
+docker compose logs -f scheduler
+docker compose logs -f worker-media
+docker compose logs -f worker-ai
+docker compose logs -f api
+```
+
+What to look for:
+
+- Scheduler logs `webdav propfind done`, `new webdav file observed`, and `webdav file stable`.
+- Worker logs `pipeline start`, `prepare media`, `transcription start`, `clip suggestion start`, `render start`, and `upload start`.
+- ffmpeg command failures include the failing command and stderr tail.
+- DeepSeek logs include transcript chunk bounds, character counts, candidate counts, and usage when returned by the SDK.
+- Thumbnail logs include frame extraction, local image generation command execution, VTuber overlay path, and final artifact path.
+- API logs include method, path, status code, and elapsed time for each frontend request.
+
+Enable SQL logs only when diagnosing database behavior:
+
+```toml
+[app]
+log_sql = true
+```
+
+This is intentionally noisy and should usually stay off in production.
+
+## Notes
+
+- FunASR/ROCm setup is intentionally outside the main Compose file for now. Benchmark the Framework host manually before binding the project to a specific GPU runtime.
+- The uploader adapter is deliberately small. `[upload].adapter = "command"` is enough to integrate a Bilibili uploader later without changing the pipeline core.
+- Existing recordings should be bootstrapped before scheduler-driven production runs, otherwise old stable files may be queued as new work.
@@ -0,0 +1,39 @@
+[alembic]
+script_location = migrations
+prepend_sys_path = .
+sqlalchemy.url = postgresql+psycopg://evanescere:evanescere@localhost:5432/evanescere
+
+[loggers]
+keys = root,sqlalchemy,alembic
+
+[handlers]
+keys = console
+
+[formatters]
+keys = generic
+
+[logger_root]
+level = WARN
+handlers = console
+qualname =
+
+[logger_sqlalchemy]
+level = WARN
+handlers =
+qualname = sqlalchemy.engine
+
+[logger_alembic]
+level = INFO
+handlers =
+qualname = alembic
+
+[handler_console]
+class = StreamHandler
+args = (sys.stderr,)
+level = NOTSET
+formatter = generic
+
+[formatter_generic]
+format = %(levelname)-5.5s [%(name)s] %(message)s
+datefmt = %H:%M:%S
+
@@ -0,0 +1,126 @@
+# Evanescere configuration file.
+#
+# Copy this file to config.toml for local Compose runs:
+#
+#   cp config.example.toml config.toml
+#
+# In containers, mount your config to:
+#
+#   /etc/evanescere/config.toml
+#
+# Secrets such as DeepSeek and WebDAV credentials belong in your private config.toml.
+
+[app]
+# Free-form environment label included in logs.
+env = "dev"
+
+# Python logging level: DEBUG, INFO, WARNING, ERROR.
+log_level = "DEBUG"
+
+# Enable SQLAlchemy SQL logging. Very noisy; keep false unless debugging DB behavior.
+log_sql = false
+
+# Frontend origins allowed to call FastAPI. Include the Vite dev server while developing.
+cors_origins = ["http://localhost:3000", "http://localhost:5173"]
+
+[database]
+# PostgreSQL connection string used by SQLAlchemy.
+# For Compose local-db profile: postgresql+psycopg://evanescere:evanescere@postgres:5432/evanescere
+# For an existing DB: postgresql+psycopg://USER:PASSWORD@HOST:5432/DBNAME
+url = "postgresql+psycopg://evanescere:evanescere@postgres:5432/evanescere"
+
+[redis]
+# Redis URL used by Dramatiq workers.
+url = "redis://redis:6379/0"
+
+[storage]
+# Container path for working media. Mount a Framework SSD directory here in Compose.
+local_root = "/data/evanescere"
+
+[webdav]
+# IIS WebDAV directory containing recording files. Point at the collection, not one file.
+base_url = "https://windows-server.example.local/webdav/recordings"
+
+# Leave username/password empty if your WebDAV endpoint does not require basic auth.
+username = ""
+password = ""
+
+# Set false only for trusted internal/self-signed testing.
+verify_tls = true
+
+# Scheduler interval. A file becomes stable after two equal positive size samples.
+poll_interval_seconds = 60
+
+[funasr]
+# FunASR-compatible API base URL. Evanescere calls /audio/transcriptions.
+base_url = "http://funasr:10096/v1"
+
+# Optional bearer token for the FunASR service.
+api_key = ""
+
+# Model name sent to FunASR. Adjust this to match your deployed service.
+model = "paraformer-zh"
+
+[deepseek]
+# OpenAI-compatible DeepSeek API base URL.
+base_url = "https://api.deepseek.com"
+
+# Required when clip suggestion is enabled.
+api_key = ""
+
+# Initial MVP model.
+model = "deepseek-v4-pro"
+
+# Lower values improve consistency.
+temperature = 0.2
+
+[defaults]
+# Initial automatic pipeline settings. These can later be changed through the API/UI.
+suggest_enabled = true
+render_enabled = true
+
+# For testing, consider false until uploads are wired and verified.
+upload_enabled = true
+
+preserve_final_artifacts = true
+bake_subtitles = true
+
+[clip]
+# LLM clip duration bounds accepted by the backend.
+min_seconds = 30
+max_seconds = 360
+
+# Transcript seconds per DeepSeek request. Larger chunks use more tokens.
+transcript_chunk_seconds = 900
+
+[thumbnail]
+# Generate thumbnails during clip render.
+enabled = true
+
+# frame_overlay: use extracted video frame as background.
+# command: run command below to call a local image generation workflow.
+provider = "frame_overlay"
+
+width = 1920
+height = 1080
+
+# Optional transparent PNG of the VTuber character.
+character_overlay_path = ""
+
+# Character height as a fraction of thumbnail height.
+character_scale = 0.42
+
+# Supported: bottom-right, bottom-left, center-right, center-left.
+character_position = "bottom-right"
+
+# Draw the clip title onto the final thumbnail.
+title_enabled = true
+
+# Used only when provider = "command". Receives JSON on stdin and must write output_path.
+command = ""
+
+[upload]
+# noop: do not upload, only record what would have happened.
+# command: run command below with JSON metadata on stdin.
+adapter = "noop"
+command = ""
@@ -0,0 +1,65 @@
+services:
+  frontend:
+    build: ./frontend
+    ports:
+      - "3000:80"
+    depends_on:
+      - api
+
+  api:
+    build: .
+    command: uvicorn evanescere.api:app --host 0.0.0.0 --port 8000
+    ports:
+      - "8000:8000"
+    volumes:
+      - ./config.toml:/etc/evanescere/config.toml:ro
+      - ./storage:/data/evanescere
+    depends_on:
+      - redis
+
+  worker-media:
+    build: .
+    command: dramatiq evanescere.jobs
+    volumes:
+      - ./config.toml:/etc/evanescere/config.toml:ro
+      - ./storage:/data/evanescere
+    depends_on:
+      - redis
+
+  worker-ai:
+    build: .
+    command: dramatiq evanescere.jobs
+    volumes:
+      - ./config.toml:/etc/evanescere/config.toml:ro
+      - ./storage:/data/evanescere
+    depends_on:
+      - redis
+
+  scheduler:
+    build: .
+    command: python -m evanescere.scheduler
+    volumes:
+      - ./config.toml:/etc/evanescere/config.toml:ro
+      - ./storage:/data/evanescere
+    depends_on:
+      - redis
+
+  redis:
+    image: redis:7-alpine
+    ports:
+      - "6379:6379"
+
+  postgres:
+    image: postgres:17-alpine
+    profiles: ["local-db"]
+    environment:
+      POSTGRES_USER: evanescere
+      POSTGRES_PASSWORD: evanescere
+      POSTGRES_DB: evanescere
+    ports:
+      - "5432:5432"
+    volumes:
+      - pgdata:/var/lib/postgresql/data
+
+volumes:
+  pgdata:
@@ -0,0 +1,5 @@
+node_modules
+dist
+.vite
+npm-debug.log
+
@@ -0,0 +1,16 @@
+FROM node:22-alpine AS build
+
+WORKDIR /app
+COPY package.json ./
+RUN npm install
+COPY index.html tsconfig.json vite.config.ts ./
+COPY public ./public
+COPY src ./src
+RUN npm run build
+
+FROM nginx:1.27-alpine
+
+COPY nginx.conf /etc/nginx/conf.d/default.conf
+COPY --from=build /app/dist /usr/share/nginx/html
+EXPOSE 80
+
@@ -0,0 +1,14 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>Evanescere</title>
+    <script src="/config.js"></script>
+  </head>
+  <body>
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>
+
@@ -0,0 +1,16 @@
+server {
+  listen 80;
+  server_name _;
+  root /usr/share/nginx/html;
+  index index.html;
+
+  location / {
+    try_files $uri $uri/ /index.html;
+  }
+
+  location = /config.js {
+    add_header Cache-Control "no-store";
+    try_files $uri =404;
+  }
+}
+
@@ -0,0 +1,23 @@
+{
+  "name": "evanescere-frontend",
+  "private": true,
+  "version": "0.1.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "tsc -b && vite build",
+    "preview": "vite preview --host 0.0.0.0"
+  },
+  "dependencies": {
+    "lucide-react": "^0.468.0",
+    "react": "^19.0.0",
+    "react-dom": "^19.0.0"
+  },
+  "devDependencies": {
+    "@types/react": "^19.0.0",
+    "@types/react-dom": "^19.0.0",
+    "@vitejs/plugin-react": "^4.3.4",
+    "typescript": "^5.7.0",
+    "vite": "^6.0.0"
+  }
+}
@@ -0,0 +1,3 @@
+window.__EVANESCERE_FRONTEND_CONFIG__ = {
+  apiBaseUrl: "http://localhost:8000"
+};
@@ -0,0 +1,304 @@
+import {
+  Check,
+  Clapperboard,
+  Play,
+  RefreshCw,
+  Save,
+  Send,
+  Settings2,
+  Upload,
+} from "lucide-react";
+import { useCallback, useEffect, useMemo, useState } from "react";
+import {
+  apiBaseUrl,
+  approveClip,
+  getClips,
+  getSettings,
+  getTranscript,
+  getVideos,
+  patchSettings,
+  renderClip,
+  runVideo,
+  uploadClip,
+} from "./api";
+import type { ClipSuggestion, PipelineSettings, TranscriptSegment, Video } from "./types";
+
+const settingLabels: Record<keyof PipelineSettings, string> = {
+  suggest_enabled: "Suggest",
+  render_enabled: "Render",
+  upload_enabled: "Upload",
+  preserve_final_artifacts: "Preserve",
+  bake_subtitles: "Bake subs",
+};
+
+function formatDuration(seconds: number | null) {
+  if (seconds === null) return "";
+  const total = Math.round(seconds);
+  const hours = Math.floor(total / 3600);
+  const minutes = Math.floor((total % 3600) / 60);
+  const secs = total % 60;
+  return hours > 0
+    ? `${hours}:${minutes.toString().padStart(2, "0")}:${secs.toString().padStart(2, "0")}`
+    : `${minutes}:${secs.toString().padStart(2, "0")}`;
+}
+
+function formatTimeRange(start: number, end: number) {
+  return `${formatDuration(start)}-${formatDuration(end)}`;
+}
+
+function statusTone(value: string) {
+  if (["done", "stable", "approved", "auto_approved"].includes(value)) return "good";
+  if (["failed", "error"].includes(value)) return "bad";
+  if (["running", "queued", "observing", "pending"].includes(value)) return "busy";
+  return "neutral";
+}
+
+export function App() {
+  const [videos, setVideos] = useState<Video[]>([]);
+  const [settings, setSettings] = useState<PipelineSettings | null>(null);
+  const [selectedVideoId, setSelectedVideoId] = useState<number | null>(null);
+  const [transcript, setTranscript] = useState<TranscriptSegment[]>([]);
+  const [clips, setClips] = useState<ClipSuggestion[]>([]);
+  const [loading, setLoading] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+
+  const selectedVideo = useMemo(
+    () => videos.find((video) => video.id === selectedVideoId) ?? null,
+    [selectedVideoId, videos],
+  );
+
+  const loadSelected = useCallback(async (videoId: number) => {
+    const [nextTranscript, nextClips] = await Promise.all([getTranscript(videoId), getClips(videoId)]);
+    setTranscript(nextTranscript);
+    setClips(nextClips);
+  }, []);
+
+  const refresh = useCallback(async () => {
+    setLoading(true);
+    setError(null);
+    try {
+      const [nextSettings, nextVideos] = await Promise.all([getSettings(), getVideos()]);
+      setSettings(nextSettings);
+      setVideos(nextVideos);
+      const targetVideoId = selectedVideoId ?? nextVideos[0]?.id ?? null;
+      setSelectedVideoId(targetVideoId);
+      if (targetVideoId !== null) {
+        await loadSelected(targetVideoId);
+      } else {
+        setTranscript([]);
+        setClips([]);
+      }
+    } catch (caught) {
+      setError(caught instanceof Error ? caught.message : "Unknown error");
+    } finally {
+      setLoading(false);
+    }
+  }, [loadSelected, selectedVideoId]);
+
+  useEffect(() => {
+    void refresh();
+  }, [refresh]);
+
+  async function handleSelect(videoId: number) {
+    setSelectedVideoId(videoId);
+    setLoading(true);
+    setError(null);
+    try {
+      await loadSelected(videoId);
+    } catch (caught) {
+      setError(caught instanceof Error ? caught.message : "Unknown error");
+    } finally {
+      setLoading(false);
+    }
+  }
+
+  async function withRefresh(action: () => Promise<unknown>) {
+    setLoading(true);
+    setError(null);
+    try {
+      await action();
+      await refresh();
+    } catch (caught) {
+      setError(caught instanceof Error ? caught.message : "Unknown error");
+      setLoading(false);
+    }
+  }
+
+  async function saveSettings() {
+    if (!settings) return;
+    await withRefresh(() => patchSettings(settings));
+  }
+
+  return (
+    <main className="app-shell">
+      <header className="topbar">
+        <div>
+          <h1>Evanescere</h1>
+          <p>{apiBaseUrl}</p>
+        </div>
+        <div className="toolbar">
+          <button type="button" className="icon-button" onClick={() => void refresh()} disabled={loading}>
+            <RefreshCw size={18} />
+            Refresh
+          </button>
+          <button type="button" className="icon-button primary" onClick={() => void saveSettings()} disabled={loading || !settings}>
+            <Save size={18} />
+            Save
+          </button>
+        </div>
+      </header>
+
+      {error && <div className="error-strip">{error}</div>}
+
+      <section className="band controls-band">
+        <div className="section-title">
+          <Settings2 size={18} />
+          <h2>Pipeline</h2>
+        </div>
+        <div className="toggle-row">
+          {settings &&
+            (Object.keys(settingLabels) as Array<keyof PipelineSettings>).map((key) => (
+              <label className="switch" key={key}>
+                <input
+                  type="checkbox"
+                  checked={settings[key]}
+                  onChange={(event) =>
+                    setSettings((current) =>
+                      current ? { ...current, [key]: event.currentTarget.checked } : current,
+                    )
+                  }
+                />
+                <span>{settingLabels[key]}</span>
+              </label>
+            ))}
+        </div>
+      </section>
+
+      <section className="workbench">
+        <aside className="video-list">
+          <div className="section-title">
+            <Clapperboard size={18} />
+            <h2>Videos</h2>
+          </div>
+          <div className="table-scroll">
+            <table>
+              <thead>
+                <tr>
+                  <th>ID</th>
+                  <th>File</th>
+                  <th>Status</th>
+                  <th></th>
+                </tr>
+              </thead>
+              <tbody>
+                {videos.map((video) => (
+                  <tr key={video.id} className={video.id === selectedVideoId ? "selected" : ""}>
+                    <td>{video.id}</td>
+                    <td>
+                      <button type="button" className="link-button" onClick={() => void handleSelect(video.id)}>
+                        {video.filename}
+                      </button>
+                      <div className="muted">{formatDuration(video.duration_sec)}</div>
+                    </td>
+                    <td>
+                      <StatusPill value={video.ingest_status} />
+                      <StatusPill value={video.processing_status} />
+                    </td>
+                    <td>
+                      <button
+                        type="button"
+                        className="square-button"
+                        title="Run pipeline"
+                        onClick={() => void withRefresh(() => runVideo(video.id))}
+                      >
+                        <Play size={17} />
+                      </button>
+                    </td>
+                  </tr>
+                ))}
+              </tbody>
+            </table>
+          </div>
+        </aside>
+
+        <section className="detail-pane">
+          <div className="detail-header">
+            <div>
+              <h2>{selectedVideo?.filename ?? "No video selected"}</h2>
+              <p>{selectedVideo?.source_url ?? ""}</p>
+            </div>
+            <div className="metric-row">
+              <Metric label="Transcript" value={transcript.length.toString()} />
+              <Metric label="Clips" value={clips.length.toString()} />
+            </div>
+          </div>
+
+          <div className="split">
+            <div className="transcript-pane">
+              <h3>Transcript</h3>
+              <div className="transcript-lines">
+                {transcript.slice(0, 180).map((segment) => (
+                  <div className="transcript-line" key={segment.id}>
+                    <span>{formatTimeRange(segment.start_sec, segment.end_sec)}</span>
+                    <p>{segment.text}</p>
+                  </div>
+                ))}
+              </div>
+            </div>
+
+            <div className="clips-pane">
+              <h3>Clips</h3>
+              <div className="clip-list">
+                {clips.map((clip) => (
+                  <article className="clip-item" key={clip.id}>
+                    <div className="clip-main">
+                      <div>
+                        <h4>{clip.title_zh}</h4>
+                        <p>{clip.summary_zh}</p>
+                      </div>
+                      <strong>{Math.round(clip.score * 100)}</strong>
+                    </div>
+                    <div className="clip-meta">
+                      <span>{formatTimeRange(clip.start_sec, clip.end_sec)}</span>
+                      <StatusPill value={clip.approval_status} />
+                      <StatusPill value={clip.render_status} />
+                      <StatusPill value={clip.upload_status} />
+                    </div>
+                    <div className="toolbar compact">
+                      <button type="button" className="icon-button" onClick={() => void withRefresh(() => approveClip(clip.id))}>
+                        <Check size={16} />
+                        Approve
+                      </button>
+                      <button type="button" className="icon-button" onClick={() => void withRefresh(() => renderClip(clip.id))}>
+                        <Send size={16} />
+                        Render
+                      </button>
+                      <button type="button" className="icon-button" onClick={() => void withRefresh(() => uploadClip(clip.id))}>
+                        <Upload size={16} />
+                        Upload
+                      </button>
+                    </div>
+                  </article>
+                ))}
+              </div>
+            </div>
+          </div>
+        </section>
+      </section>
+    </main>
+  );
+}
+
+function StatusPill({ value }: { value: string }) {
+  return <span className={`status-pill ${statusTone(value)}`}>{value}</span>;
+}
+
+function Metric({ label, value }: { label: string; value: string }) {
+  return (
+    <div className="metric">
+      <span>{label}</span>
+      <strong>{value}</strong>
+    </div>
+  );
+}
+
@@ -0,0 +1,66 @@
+import type { Artifact, ClipSuggestion, PipelineSettings, TranscriptSegment, Video } from "./types";
+
+const runtimeApiBase = window.__EVANESCERE_FRONTEND_CONFIG__?.apiBaseUrl;
+
+export const apiBaseUrl = (runtimeApiBase || "http://localhost:8000").replace(
+  /\/$/,
+  "",
+);
+
+async function request<T>(path: string, init: RequestInit = {}): Promise<T> {
+  const response = await fetch(`${apiBaseUrl}${path}`, {
+    ...init,
+    headers: {
+      "Content-Type": "application/json",
+      ...init.headers,
+    },
+  });
+  if (!response.ok) {
+    const text = await response.text();
+    throw new Error(text || `${response.status} ${response.statusText}`);
+  }
+  return response.json() as Promise<T>;
+}
+
+export function getSettings(): Promise<PipelineSettings> {
+  return request<PipelineSettings>("/settings");
+}
+
+export function patchSettings(settings: PipelineSettings): Promise<PipelineSettings> {
+  return request<PipelineSettings>("/settings", {
+    method: "PATCH",
+    body: JSON.stringify(settings),
+  });
+}
+
+export function getVideos(): Promise<Video[]> {
+  return request<Video[]>("/videos");
+}
+
+export function runVideo(videoId: number) {
+  return request(`/videos/${videoId}/run`, { method: "POST" });
+}
+
+export function getTranscript(videoId: number): Promise<TranscriptSegment[]> {
+  return request<TranscriptSegment[]>(`/videos/${videoId}/transcript`);
+}
+
+export function getClips(videoId: number): Promise<ClipSuggestion[]> {
+  return request<ClipSuggestion[]>(`/videos/${videoId}/clips`);
+}
+
+export function getClipArtifacts(clipId: number): Promise<Artifact[]> {
+  return request<Artifact[]>(`/clips/${clipId}/artifacts`);
+}
+
+export function approveClip(clipId: number): Promise<ClipSuggestion> {
+  return request<ClipSuggestion>(`/clips/${clipId}/approve`, { method: "POST" });
+}
+
+export function renderClip(clipId: number): Promise<ClipSuggestion> {
+  return request<ClipSuggestion>(`/clips/${clipId}/render`, { method: "POST" });
+}
+
+export function uploadClip(clipId: number): Promise<ClipSuggestion> {
+  return request<ClipSuggestion>(`/clips/${clipId}/upload`, { method: "POST" });
+}
@@ -0,0 +1,11 @@
+import { StrictMode } from "react";
+import { createRoot } from "react-dom/client";
+import { App } from "./App";
+import "./styles.css";
+
+createRoot(document.getElementById("root")!).render(
+  <StrictMode>
+    <App />
+  </StrictMode>,
+);
+
@@ -0,0 +1,434 @@
+* {
+  box-sizing: border-box;
+}
+
+:root {
+  color: #1f252b;
+  background: #f4f6f8;
+  font-family:
+    Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+  font-synthesis: none;
+  text-rendering: optimizeLegibility;
+}
+
+body {
+  margin: 0;
+  min-width: 320px;
+  min-height: 100vh;
+  background: #f4f6f8;
+}
+
+button,
+input {
+  font: inherit;
+}
+
+button {
+  white-space: nowrap;
+}
+
+.app-shell {
+  width: min(1440px, 100%);
+  margin: 0 auto;
+  padding: 20px;
+}
+
+.topbar {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+  padding: 8px 0 20px;
+}
+
+.topbar h1 {
+  margin: 0;
+  color: #18222b;
+  font-size: 28px;
+  font-weight: 720;
+}
+
+.topbar p,
+.detail-header p,
+.muted {
+  margin: 4px 0 0;
+  color: #66737f;
+  font-size: 13px;
+  overflow-wrap: anywhere;
+}
+
+.toolbar {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+}
+
+.toolbar.compact {
+  margin-top: 12px;
+}
+
+.icon-button,
+.square-button {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 7px;
+  min-height: 36px;
+  border: 1px solid #b7c0ca;
+  border-radius: 6px;
+  background: #ffffff;
+  color: #1f252b;
+  cursor: pointer;
+}
+
+.icon-button {
+  padding: 0 12px;
+}
+
+.square-button {
+  width: 36px;
+  padding: 0;
+}
+
+.icon-button:hover,
+.square-button:hover {
+  border-color: #21756b;
+  background: #e9f3f1;
+}
+
+.icon-button.primary {
+  border-color: #21756b;
+  background: #21756b;
+  color: #ffffff;
+}
+
+.icon-button:disabled,
+.square-button:disabled {
+  cursor: progress;
+  opacity: 0.6;
+}
+
+.error-strip {
+  margin-bottom: 14px;
+  border-left: 4px solid #b42318;
+  background: #fff3f0;
+  color: #76180f;
+  padding: 10px 12px;
+  border-radius: 6px;
+  font-size: 14px;
+  overflow-wrap: anywhere;
+}
+
+.band {
+  border-top: 1px solid #d8dee6;
+  padding: 16px 0;
+}
+
+.controls-band {
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  gap: 18px;
+}
+
+.section-title {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  color: #2f3b46;
+}
+
+.section-title h2,
+.detail-header h2,
+.transcript-pane h3,
+.clips-pane h3 {
+  margin: 0;
+  font-size: 16px;
+  font-weight: 690;
+}
+
+.toggle-row {
+  display: flex;
+  gap: 10px;
+  flex-wrap: wrap;
+}
+
+.switch {
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  min-height: 34px;
+  padding: 0 10px;
+  border: 1px solid #ccd3db;
+  border-radius: 6px;
+  background: #ffffff;
+  font-size: 13px;
+  color: #2f3b46;
+}
+
+.switch input {
+  width: 16px;
+  height: 16px;
+  accent-color: #21756b;
+}
+
+.workbench {
+  display: grid;
+  grid-template-columns: minmax(340px, 0.36fr) minmax(0, 1fr);
+  gap: 18px;
+  align-items: start;
+}
+
+.video-list,
+.detail-pane {
+  min-width: 0;
+}
+
+.video-list {
+  border-top: 1px solid #d8dee6;
+  padding-top: 16px;
+}
+
+.table-scroll {
+  margin-top: 12px;
+  overflow: auto;
+  background: #ffffff;
+  border: 1px solid #d8dee6;
+  border-radius: 8px;
+}
+
+table {
+  width: 100%;
+  min-width: 520px;
+  border-collapse: collapse;
+  font-size: 13px;
+}
+
+th,
+td {
+  padding: 10px;
+  border-bottom: 1px solid #e4e8ed;
+  text-align: left;
+  vertical-align: top;
+}
+
+th {
+  color: #586674;
+  font-weight: 650;
+  background: #fbfcfd;
+}
+
+tr.selected td {
+  background: #eef6f5;
+}
+
+.link-button {
+  display: inline;
+  border: 0;
+  padding: 0;
+  background: transparent;
+  color: #174b91;
+  cursor: pointer;
+  text-align: left;
+  overflow-wrap: anywhere;
+}
+
+.detail-pane {
+  border-top: 1px solid #d8dee6;
+  padding-top: 16px;
+}
+
+.detail-header {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 16px;
+  margin-bottom: 14px;
+}
+
+.metric-row {
+  display: flex;
+  gap: 10px;
+}
+
+.metric {
+  min-width: 88px;
+  border: 1px solid #d8dee6;
+  border-radius: 8px;
+  background: #ffffff;
+  padding: 8px 10px;
+}
+
+.metric span {
+  display: block;
+  color: #66737f;
+  font-size: 12px;
+}
+
+.metric strong {
+  display: block;
+  margin-top: 3px;
+  font-size: 20px;
+  font-weight: 720;
+}
+
+.split {
+  display: grid;
+  grid-template-columns: minmax(280px, 0.95fr) minmax(320px, 1.05fr);
+  gap: 18px;
+}
+
+.transcript-pane,
+.clips-pane {
+  min-width: 0;
+}
+
+.transcript-lines,
+.clip-list {
+  margin-top: 12px;
+  max-height: 68vh;
+  overflow: auto;
+}
+
+.transcript-lines {
+  border: 1px solid #d8dee6;
+  border-radius: 8px;
+  background: #ffffff;
+}
+
+.transcript-line {
+  display: grid;
+  grid-template-columns: 112px minmax(0, 1fr);
+  gap: 10px;
+  padding: 9px 10px;
+  border-bottom: 1px solid #edf0f3;
+}
+
+.transcript-line span {
+  color: #66737f;
+  font-size: 12px;
+  font-variant-numeric: tabular-nums;
+}
+
+.transcript-line p {
+  margin: 0;
+  overflow-wrap: anywhere;
+  line-height: 1.45;
+}
+
+.clip-list {
+  display: grid;
+  gap: 10px;
+}
+
+.clip-item {
+  border: 1px solid #d8dee6;
+  border-radius: 8px;
+  background: #ffffff;
+  padding: 12px;
+}
+
+.clip-main {
+  display: flex;
+  justify-content: space-between;
+  gap: 12px;
+}
+
+.clip-main h4 {
+  margin: 0;
+  color: #18222b;
+  font-size: 15px;
+}
+
+.clip-main p {
+  margin: 5px 0 0;
+  color: #4f5c68;
+  line-height: 1.45;
+}
+
+.clip-main strong {
+  display: grid;
+  place-items: center;
+  flex: 0 0 44px;
+  width: 44px;
+  height: 44px;
+  border-radius: 50%;
+  background: #eef6f5;
+  color: #155c54;
+}
+
+.clip-meta {
+  display: flex;
+  align-items: center;
+  gap: 7px;
+  flex-wrap: wrap;
+  margin-top: 10px;
+  color: #66737f;
+  font-size: 12px;
+}
+
+.status-pill {
+  display: inline-flex;
+  align-items: center;
+  min-height: 22px;
+  padding: 0 7px;
+  border-radius: 999px;
+  font-size: 12px;
+  font-variant-numeric: tabular-nums;
+  background: #edf0f3;
+  color: #4f5c68;
+}
+
+.status-pill.good {
+  background: #e5f5ec;
+  color: #17623b;
+}
+
+.status-pill.bad {
+  background: #fff0ec;
+  color: #a02717;
+}
+
+.status-pill.busy {
+  background: #fff4d6;
+  color: #6f4d00;
+}
+
+@media (max-width: 1100px) {
+  .workbench,
+  .split {
+    grid-template-columns: 1fr;
+  }
+
+  .transcript-lines,
+  .clip-list {
+    max-height: none;
+  }
+}
+
+@media (max-width: 720px) {
+  .app-shell {
+    padding: 14px;
+  }
+
+  .topbar,
+  .controls-band,
+  .detail-header {
+    align-items: stretch;
+    flex-direction: column;
+  }
+
+  .metric-row {
+    width: 100%;
+  }
+
+  .metric {
+    flex: 1;
+  }
+
+  .transcript-line {
+    grid-template-columns: 1fr;
+  }
+}
+
@@ -0,0 +1,61 @@
+export interface Video {
+  id: number;
+  source_url: string;
+  filename: string;
+  size_bytes: number | null;
+  duration_sec: number | null;
+  codec_metadata: Record<string, unknown>;
+  ingest_status: string;
+  processing_status: string;
+  created_at: string;
+  updated_at: string;
+}
+
+export interface TranscriptSegment {
+  id: number;
+  video_id: number;
+  start_sec: number;
+  end_sec: number;
+  text: string;
+  speaker: string | null;
+  confidence: number | null;
+  segment_metadata: Record<string, unknown>;
+}
+
+export interface ClipSuggestion {
+  id: number;
+  video_id: number;
+  start_sec: number;
+  end_sec: number;
+  title_zh: string;
+  summary_zh: string;
+  reason: string;
+  score: number;
+  tags: string[];
+  subtitle_priority: string;
+  approval_status: string;
+  render_status: string;
+  upload_status: string;
+  created_at: string;
+  updated_at: string;
+}
+
+export interface PipelineSettings {
+  suggest_enabled: boolean;
+  render_enabled: boolean;
+  upload_enabled: boolean;
+  preserve_final_artifacts: boolean;
+  bake_subtitles: boolean;
+}
+
+export interface Artifact {
+  id: number;
+  video_id: number | null;
+  clip_id: number | null;
+  artifact_type: string;
+  local_path: string;
+  webdav_url: string | null;
+  preserve: boolean;
+  artifact_metadata: Record<string, unknown>;
+  created_at: string;
+}
@@ -0,0 +1,7 @@
+/// <reference types="vite/client" />
+
+interface Window {
+  __EVANESCERE_FRONTEND_CONFIG__?: {
+    apiBaseUrl?: string;
+  };
+}
@@ -0,0 +1,22 @@
+{
+  "compilerOptions": {
+    "composite": true,
+    "target": "ES2022",
+    "useDefineForClassFields": true,
+    "lib": ["ES2022", "DOM", "DOM.Iterable"],
+    "allowJs": false,
+    "skipLibCheck": true,
+    "esModuleInterop": true,
+    "allowSyntheticDefaultImports": true,
+    "strict": true,
+    "forceConsistentCasingInFileNames": true,
+    "module": "ESNext",
+    "moduleResolution": "Bundler",
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "noEmit": true,
+    "jsx": "react-jsx"
+  },
+  "include": ["src"]
+}
+
@@ -0,0 +1,11 @@
+import { defineConfig } from "vite";
+import react from "@vitejs/plugin-react";
+
+export default defineConfig({
+  plugins: [react()],
+  server: {
+    host: "0.0.0.0",
+    port: 5173,
+  },
+});
+
@@ -0,0 +1,44 @@
+from logging.config import fileConfig
+
+from alembic import context
+from sqlalchemy import engine_from_config, pool
+
+from evanescere.config import get_settings
+from evanescere.models import Base
+
+config = context.config
+
+if config.config_file_name is not None:
+    fileConfig(config.config_file_name)
+
+config.set_main_option("sqlalchemy.url", get_settings().database_url)
+target_metadata = Base.metadata
+
+
+def run_migrations_offline() -> None:
+    url = config.get_main_option("sqlalchemy.url")
+    context.configure(url=url, target_metadata=target_metadata, literal_binds=True)
+
+    with context.begin_transaction():
+        context.run_migrations()
+
+
+def run_migrations_online() -> None:
+    connectable = engine_from_config(
+        config.get_section(config.config_ini_section, {}),
+        prefix="sqlalchemy.",
+        poolclass=pool.NullPool,
+    )
+
+    with connectable.connect() as connection:
+        context.configure(connection=connection, target_metadata=target_metadata)
+
+        with context.begin_transaction():
+            context.run_migrations()
+
+
+if context.is_offline_mode():
+    run_migrations_offline()
+else:
+    run_migrations_online()
+
@@ -0,0 +1,113 @@
+"""Initial schema.
+
+Revision ID: 0001_initial
+Revises:
+Create Date: 2026-05-29
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+revision: str = "0001_initial"
+down_revision: str | None = None
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "videos",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("source_url", sa.Text(), nullable=False, unique=True),
+        sa.Column("filename", sa.Text(), nullable=False),
+        sa.Column("size_bytes", sa.BigInteger()),
+        sa.Column("size_samples", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("duration_sec", sa.Float()),
+        sa.Column("codec_metadata", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("ingest_status", sa.String(length=32), nullable=False),
+        sa.Column("processing_status", sa.String(length=32), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
+    )
+    op.create_index("ix_videos_ingest_status", "videos", ["ingest_status"])
+
+    op.create_table(
+        "pipeline_runs",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("video_id", sa.Integer(), sa.ForeignKey("videos.id", ondelete="CASCADE"), nullable=False),
+        sa.Column("trigger", sa.String(length=32), nullable=False),
+        sa.Column("stage", sa.String(length=64), nullable=False),
+        sa.Column("status", sa.String(length=32), nullable=False),
+        sa.Column("error", sa.Text()),
+        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
+    )
+
+    op.create_table(
+        "artifacts",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("video_id", sa.Integer(), sa.ForeignKey("videos.id", ondelete="CASCADE")),
+        sa.Column("clip_id", sa.Integer()),
+        sa.Column("artifact_type", sa.String(length=64), nullable=False),
+        sa.Column("local_path", sa.Text(), nullable=False),
+        sa.Column("webdav_url", sa.Text()),
+        sa.Column("preserve", sa.Boolean(), nullable=False),
+        sa.Column("metadata", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
+    )
+
+    op.create_table(
+        "transcript_segments",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("video_id", sa.Integer(), sa.ForeignKey("videos.id", ondelete="CASCADE"), nullable=False),
+        sa.Column("start_sec", sa.Float(), nullable=False),
+        sa.Column("end_sec", sa.Float(), nullable=False),
+        sa.Column("text", sa.Text(), nullable=False),
+        sa.Column("speaker", sa.Text()),
+        sa.Column("confidence", sa.Float()),
+        sa.Column("metadata", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
+    )
+    op.create_index("ix_transcript_segments_video_id", "transcript_segments", ["video_id"])
+
+    op.create_table(
+        "clip_suggestions",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("video_id", sa.Integer(), sa.ForeignKey("videos.id", ondelete="CASCADE"), nullable=False),
+        sa.Column("start_sec", sa.Float(), nullable=False),
+        sa.Column("end_sec", sa.Float(), nullable=False),
+        sa.Column("title_zh", sa.Text(), nullable=False),
+        sa.Column("summary_zh", sa.Text(), nullable=False),
+        sa.Column("reason", sa.Text(), nullable=False),
+        sa.Column("score", sa.Float(), nullable=False),
+        sa.Column("tags", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("subtitle_priority", sa.String(length=32), nullable=False),
+        sa.Column("llm_raw", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("approval_status", sa.String(length=32), nullable=False),
+        sa.Column("render_status", sa.String(length=32), nullable=False),
+        sa.Column("upload_status", sa.String(length=32), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
+    )
+
+    op.create_table(
+        "settings",
+        sa.Column("key", sa.String(length=128), primary_key=True),
+        sa.Column("value", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
+    )
+
+
+def downgrade() -> None:
+    op.drop_table("settings")
+    op.drop_table("clip_suggestions")
+    op.drop_index("ix_transcript_segments_video_id", table_name="transcript_segments")
+    op.drop_table("transcript_segments")
+    op.drop_table("artifacts")
+    op.drop_table("pipeline_runs")
+    op.drop_index("ix_videos_ingest_status", table_name="videos")
+    op.drop_table("videos")
+
@@ -0,0 +1,42 @@
+[project]
+name = "evanescere"
+version = "0.1.0"
+description = "Livestream transcription, clip suggestion, rendering, and upload orchestration."
+requires-python = ">=3.12"
+dependencies = [
+  "alembic>=1.13",
+  "dramatiq[redis]>=1.17",
+  "fastapi>=0.115",
+  "httpx>=0.27",
+  "openai>=1.55",
+  "pillow>=11.0",
+  "psycopg[binary]>=3.2",
+  "pydantic>=2.9",
+  "python-multipart>=0.0.12",
+  "sqlalchemy>=2.0",
+  "typer>=0.12",
+  "uvicorn[standard]>=0.32",
+]
+
+[project.optional-dependencies]
+dev = [
+  "pytest>=8.3",
+  "ruff>=0.8",
+]
+
+[project.scripts]
+evanescere = "evanescere.cli:app"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/evanescere"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py312"
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "UP", "B"]
@@ -0,0 +1,4 @@
+__all__ = ["__version__"]
+
+__version__ = "0.1.0"
+
@@ -0,0 +1,180 @@
+import logging
+import time
+
+from fastapi import Depends, FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi import Request
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from evanescere import __version__
+from evanescere.config import get_settings
+from evanescere.db import get_db
+from evanescere.jobs import enqueue_pipeline, enqueue_render, enqueue_upload
+from evanescere.logging_config import configure_logging
+from evanescere.models import Artifact, ClipSuggestion, PipelineRun, TranscriptSegment, Video
+from evanescere.schemas import (
+    ArtifactRead,
+    ClipSuggestionRead,
+    RunRead,
+    SettingsPatch,
+    SettingsRead,
+    TranscriptSegmentRead,
+    VideoRead,
+)
+from evanescere.settings_store import get_pipeline_settings, patch_pipeline_settings
+
+configure_logging()
+app = FastAPI(title="Evanescere", version=__version__)
+logger = logging.getLogger(__name__)
+
+settings = get_settings()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=settings.cors_origin_list,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+@app.middleware("http")
+async def log_requests(request: Request, call_next):
+    start = time.perf_counter()
+    response = await call_next(request)
+    elapsed_ms = (time.perf_counter() - start) * 1000
+    logger.info(
+        "api request method=%s path=%s status=%s elapsed_ms=%.1f",
+        request.method,
+        request.url.path,
+        response.status_code,
+        elapsed_ms,
+    )
+    return response
+
+
+@app.get("/")
+def root() -> dict[str, str]:
+    return {"service": "evanescere-api", "version": __version__}
+
+
+@app.get("/health")
+def health() -> dict[str, str]:
+    return {"status": "ok", "version": __version__}
+
+
+@app.get("/videos", response_model=list[VideoRead])
+def list_videos(db: Session = Depends(get_db)) -> list[Video]:
+    return list(db.scalars(select(Video).order_by(Video.created_at.desc())).all())
+
+
+@app.post("/videos/{video_id}/run", response_model=RunRead)
+def run_video(video_id: int, db: Session = Depends(get_db)) -> PipelineRun:
+    video = db.get(Video, video_id)
+    if video is None:
+        raise HTTPException(status_code=404, detail="Video not found")
+    run = PipelineRun(video_id=video_id, trigger="manual", stage="queued", status="queued")
+    video.processing_status = "queued"
+    db.add(run)
+    db.flush()
+    enqueue_pipeline(video_id, run.id)
+    return run
+
+
+@app.get("/videos/{video_id}/transcript", response_model=list[TranscriptSegmentRead])
+def video_transcript(video_id: int, db: Session = Depends(get_db)) -> list[TranscriptSegment]:
+    return list(
+        db.scalars(
+            select(TranscriptSegment)
+            .where(TranscriptSegment.video_id == video_id)
+            .order_by(TranscriptSegment.start_sec)
+        ).all()
+    )
+
+
+@app.get("/videos/{video_id}/clips", response_model=list[ClipSuggestionRead])
+def video_clips(video_id: int, db: Session = Depends(get_db)) -> list[ClipSuggestion]:
+    return list(
+        db.scalars(
+            select(ClipSuggestion)
+            .where(ClipSuggestion.video_id == video_id)
+            .order_by(ClipSuggestion.score.desc(), ClipSuggestion.start_sec)
+        ).all()
+    )
+
+
+@app.get("/videos/{video_id}/artifacts", response_model=list[ArtifactRead])
+def video_artifacts(video_id: int, db: Session = Depends(get_db)) -> list[Artifact]:
+    return list(
+        db.scalars(
+            select(Artifact)
+            .where(Artifact.video_id == video_id)
+            .order_by(Artifact.created_at.desc())
+        ).all()
+    )
+
+
+@app.get("/clips/{clip_id}/artifacts", response_model=list[ArtifactRead])
+def clip_artifacts(clip_id: int, db: Session = Depends(get_db)) -> list[Artifact]:
+    return list(
+        db.scalars(
+            select(Artifact)
+            .where(Artifact.clip_id == clip_id)
+            .order_by(Artifact.created_at.desc())
+        ).all()
+    )
+
+
+@app.post("/clips/{clip_id}/approve", response_model=ClipSuggestionRead)
+def approve_clip(clip_id: int, db: Session = Depends(get_db)) -> ClipSuggestion:
+    clip = db.get(ClipSuggestion, clip_id)
+    if clip is None:
+        raise HTTPException(status_code=404, detail="Clip not found")
+    clip.approval_status = "approved"
+    return clip
+
+
+@app.post("/clips/{clip_id}/render", response_model=ClipSuggestionRead)
+def render_clip(clip_id: int, db: Session = Depends(get_db)) -> ClipSuggestion:
+    clip = db.get(ClipSuggestion, clip_id)
+    if clip is None:
+        raise HTTPException(status_code=404, detail="Clip not found")
+    clip.render_status = "queued"
+    enqueue_render(clip_id)
+    return clip
+
+
+@app.post("/clips/{clip_id}/upload", response_model=ClipSuggestionRead)
+def upload_clip(clip_id: int, db: Session = Depends(get_db)) -> ClipSuggestion:
+    clip = db.get(ClipSuggestion, clip_id)
+    if clip is None:
+        raise HTTPException(status_code=404, detail="Clip not found")
+    clip.upload_status = "queued"
+    enqueue_upload(clip_id)
+    return clip
+
+
+@app.get("/runs/{run_id}", response_model=RunRead)
+def get_run(run_id: int, db: Session = Depends(get_db)) -> PipelineRun:
+    run = db.get(PipelineRun, run_id)
+    if run is None:
+        raise HTTPException(status_code=404, detail="Run not found")
+    return run
+
+
+@app.get("/artifacts/{artifact_id}", response_model=ArtifactRead)
+def get_artifact(artifact_id: int, db: Session = Depends(get_db)) -> Artifact:
+    artifact = db.get(Artifact, artifact_id)
+    if artifact is None:
+        raise HTTPException(status_code=404, detail="Artifact not found")
+    return artifact
+
+
+@app.get("/settings", response_model=SettingsRead)
+def read_settings(db: Session = Depends(get_db)) -> SettingsRead:
+    return get_pipeline_settings(db)
+
+
+@app.patch("/settings", response_model=SettingsRead)
+def update_settings(patch: SettingsPatch, db: Session = Depends(get_db)) -> SettingsRead:
+    return patch_pipeline_settings(db, patch)
@@ -0,0 +1,75 @@
+import typer
+from sqlalchemy import select
+
+from evanescere.db import session_scope
+from evanescere.jobs import enqueue_pipeline, enqueue_render, enqueue_transcribe, enqueue_suggest, enqueue_upload
+from evanescere.logging_config import configure_logging
+from evanescere.models import PipelineRun, Video
+from evanescere.services.webdav import WebDavClient, bootstrap_existing as bootstrap_webdav_existing, scan_once
+
+app = typer.Typer(no_args_is_help=True)
+configure_logging()
+
+
+@app.command()
+def bootstrap_existing() -> None:
+    with session_scope() as session:
+        count = bootstrap_webdav_existing(WebDavClient.from_settings(), session)
+    typer.echo(f"Marked {count} existing WebDAV recordings as existing_done.")
+
+
+@app.command()
+def scan() -> None:
+    with session_scope() as session:
+        observed, stable = scan_once(WebDavClient.from_settings(), session)
+    typer.echo(f"Observed {observed} files; {stable} newly stable.")
+
+
+@app.command("run-video")
+def run_video(video_id: int) -> None:
+    with session_scope() as session:
+        video = session.get(Video, video_id)
+        if video is None:
+            raise typer.BadParameter("Video not found")
+        run = PipelineRun(video_id=video_id, trigger="manual", stage="queued", status="queued")
+        video.processing_status = "queued"
+        session.add(run)
+        session.flush()
+        enqueue_pipeline(video_id, run.id)
+    typer.echo(f"Queued pipeline for video {video_id}.")
+
+
+@app.command()
+def transcribe(video_id: int) -> None:
+    enqueue_transcribe(video_id)
+    typer.echo(f"Queued transcription for video {video_id}.")
+
+
+@app.command()
+def suggest(video_id: int) -> None:
+    enqueue_suggest(video_id)
+    typer.echo(f"Queued clip suggestion for video {video_id}.")
+
+
+@app.command()
+def render(clip_id: int) -> None:
+    enqueue_render(clip_id)
+    typer.echo(f"Queued render for clip {clip_id}.")
+
+
+@app.command()
+def upload(clip_id: int) -> None:
+    enqueue_upload(clip_id)
+    typer.echo(f"Queued upload for clip {clip_id}.")
+
+
+@app.command()
+def videos() -> None:
+    with session_scope() as session:
+        rows = session.scalars(select(Video).order_by(Video.created_at.desc())).all()
+        for video in rows:
+            typer.echo(f"{video.id}\t{video.ingest_status}\t{video.processing_status}\t{video.filename}")
+
+
+if __name__ == "__main__":
+    app()
@@ -0,0 +1,150 @@
+from __future__ import annotations
+
+import tomllib
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+from pydantic import AnyHttpUrl, BaseModel, Field
+
+DEFAULT_CONFIG_PATHS = (
+    Path("/etc/evanescere/config.toml"),
+    Path("config.toml"),
+)
+
+
+class Settings(BaseModel):
+    app_env: str = "dev"
+    log_level: str = "INFO"
+    log_sql: bool = False
+    cors_origins: list[str] = Field(default_factory=lambda: ["http://localhost:3000", "http://localhost:5173"])
+
+    database_url: str = "postgresql+psycopg://evanescere:evanescere@localhost:5432/evanescere"
+    redis_url: str = "redis://localhost:6379/0"
+
+    local_storage_root: Path = Path("./storage")
+
+    webdav_base_url: AnyHttpUrl | str = "http://localhost/webdav/recordings"
+    webdav_username: str | None = None
+    webdav_password: str | None = None
+    webdav_verify_tls: bool = True
+    webdav_poll_interval_seconds: int = Field(default=60, ge=5)
+
+    funasr_base_url: str = "http://localhost:10096/v1"
+    funasr_api_key: str | None = None
+    funasr_model: str = "paraformer-zh"
+
+    deepseek_base_url: str = "https://api.deepseek.com"
+    deepseek_api_key: str | None = None
+    deepseek_model: str = "deepseek-v4-pro"
+    deepseek_temperature: float = 0.2
+
+    default_suggest_enabled: bool = True
+    default_render_enabled: bool = True
+    default_upload_enabled: bool = True
+    default_preserve_final_artifacts: bool = True
+    default_bake_subtitles: bool = True
+
+    upload_adapter: str = "noop"
+    upload_command: str | None = None
+
+    clip_min_seconds: int = 30
+    clip_max_seconds: int = 360
+    transcript_chunk_seconds: int = 900
+
+    thumbnail_enabled: bool = True
+    thumbnail_provider: str = "frame_overlay"
+    thumbnail_width: int = Field(default=1920, ge=320)
+    thumbnail_height: int = Field(default=1080, ge=180)
+    thumbnail_character_overlay_path: str | None = None
+    thumbnail_character_scale: float = Field(default=0.42, gt=0, le=1)
+    thumbnail_character_position: str = "bottom-right"
+    thumbnail_title_enabled: bool = True
+    thumbnail_command: str | None = None
+
+    @property
+    def cors_origin_list(self) -> list[str]:
+        return self.cors_origins
+
+
+def configured_path() -> Path | None:
+    for path in DEFAULT_CONFIG_PATHS:
+        if path.exists():
+            return path
+    return None
+
+
+def load_toml_settings(path: Path) -> dict[str, Any]:
+    with path.open("rb") as config_file:
+        raw = tomllib.load(config_file)
+    return flatten_config(raw)
+
+
+def blank_to_none(value: Any) -> Any:
+    if value == "":
+        return None
+    return value
+
+
+def flatten_config(raw: dict[str, Any]) -> dict[str, Any]:
+    app = raw.get("app", {})
+    database = raw.get("database", {})
+    redis = raw.get("redis", {})
+    storage = raw.get("storage", {})
+    webdav = raw.get("webdav", {})
+    funasr = raw.get("funasr", {})
+    deepseek = raw.get("deepseek", {})
+    defaults = raw.get("defaults", {})
+    clip = raw.get("clip", {})
+    thumbnail = raw.get("thumbnail", {})
+    upload = raw.get("upload", {})
+
+    return {
+        "app_env": app.get("env", "dev"),
+        "log_level": app.get("log_level", "INFO"),
+        "log_sql": app.get("log_sql", False),
+        "cors_origins": app.get("cors_origins", ["http://localhost:3000", "http://localhost:5173"]),
+        "database_url": database.get("url", Settings.model_fields["database_url"].default),
+        "redis_url": redis.get("url", Settings.model_fields["redis_url"].default),
+        "local_storage_root": storage.get("local_root", Settings.model_fields["local_storage_root"].default),
+        "webdav_base_url": webdav.get("base_url", Settings.model_fields["webdav_base_url"].default),
+        "webdav_username": blank_to_none(webdav.get("username")),
+        "webdav_password": blank_to_none(webdav.get("password")),
+        "webdav_verify_tls": webdav.get("verify_tls", True),
+        "webdav_poll_interval_seconds": webdav.get("poll_interval_seconds", 60),
+        "funasr_base_url": funasr.get("base_url", Settings.model_fields["funasr_base_url"].default),
+        "funasr_api_key": blank_to_none(funasr.get("api_key")),
+        "funasr_model": funasr.get("model", "paraformer-zh"),
+        "deepseek_base_url": deepseek.get("base_url", "https://api.deepseek.com"),
+        "deepseek_api_key": blank_to_none(deepseek.get("api_key")),
+        "deepseek_model": deepseek.get("model", "deepseek-v4-pro"),
+        "deepseek_temperature": deepseek.get("temperature", 0.2),
+        "default_suggest_enabled": defaults.get("suggest_enabled", True),
+        "default_render_enabled": defaults.get("render_enabled", True),
+        "default_upload_enabled": defaults.get("upload_enabled", True),
+        "default_preserve_final_artifacts": defaults.get("preserve_final_artifacts", True),
+        "default_bake_subtitles": defaults.get("bake_subtitles", True),
+        "clip_min_seconds": clip.get("min_seconds", 30),
+        "clip_max_seconds": clip.get("max_seconds", 360),
+        "transcript_chunk_seconds": clip.get("transcript_chunk_seconds", 900),
+        "thumbnail_enabled": thumbnail.get("enabled", True),
+        "thumbnail_provider": thumbnail.get("provider", "frame_overlay"),
+        "thumbnail_width": thumbnail.get("width", 1920),
+        "thumbnail_height": thumbnail.get("height", 1080),
+        "thumbnail_character_overlay_path": blank_to_none(thumbnail.get("character_overlay_path")),
+        "thumbnail_character_scale": thumbnail.get("character_scale", 0.42),
+        "thumbnail_character_position": thumbnail.get("character_position", "bottom-right"),
+        "thumbnail_title_enabled": thumbnail.get("title_enabled", True),
+        "thumbnail_command": blank_to_none(thumbnail.get("command")),
+        "upload_adapter": upload.get("adapter", "noop"),
+        "upload_command": blank_to_none(upload.get("command")),
+    }
+
+
+@lru_cache
+def get_settings() -> Settings:
+    path = configured_path()
+    values = load_toml_settings(path) if path else {}
+    settings = Settings(**values)
+    settings.local_storage_root.mkdir(parents=True, exist_ok=True)
+    return settings
@@ -0,0 +1,29 @@
+from collections.abc import Iterator
+from contextlib import contextmanager
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session, sessionmaker
+
+from evanescere.config import get_settings
+
+settings = get_settings()
+engine = create_engine(settings.database_url, pool_pre_ping=True, echo=settings.log_sql)
+SessionLocal = sessionmaker(bind=engine, autoflush=False, expire_on_commit=False)
+
+
+@contextmanager
+def session_scope() -> Iterator[Session]:
+    session = SessionLocal()
+    try:
+        yield session
+        session.commit()
+    except Exception:
+        session.rollback()
+        raise
+    finally:
+        session.close()
+
+
+def get_db() -> Iterator[Session]:
+    with session_scope() as session:
+        yield session
@@ -0,0 +1,87 @@
+import logging
+
+import dramatiq
+from dramatiq.brokers.redis import RedisBroker
+
+from evanescere.config import get_settings
+from evanescere.db import session_scope
+from evanescere.logging_config import configure_logging
+from evanescere.models import Video
+from evanescere.pipeline import render_clip_by_id, run_pipeline, suggest_clips, transcribe_video, upload_clip_by_id
+
+configure_logging()
+logger = logging.getLogger(__name__)
+
+redis_broker = RedisBroker(url=get_settings().redis_url)
+dramatiq.set_broker(redis_broker)
+
+
+def enqueue_pipeline(video_id: int, run_id: int | None = None) -> None:
+    logger.info("enqueue pipeline video_id=%s run_id=%s", video_id, run_id)
+    pipeline_job.send(video_id, run_id)
+
+
+def enqueue_transcribe(video_id: int) -> None:
+    logger.info("enqueue transcribe video_id=%s", video_id)
+    transcribe_job.send(video_id)
+
+
+def enqueue_suggest(video_id: int) -> None:
+    logger.info("enqueue suggest video_id=%s", video_id)
+    suggest_job.send(video_id)
+
+
+def enqueue_render(clip_id: int) -> None:
+    logger.info("enqueue render clip_id=%s", clip_id)
+    render_job.send(clip_id)
+
+
+def enqueue_upload(clip_id: int) -> None:
+    logger.info("enqueue upload clip_id=%s", clip_id)
+    upload_job.send(clip_id)
+
+
+@dramatiq.actor(max_retries=1)
+def pipeline_job(video_id: int, run_id: int | None = None) -> None:
+    logger.info("job start pipeline video_id=%s run_id=%s", video_id, run_id)
+    with session_scope() as session:
+        run_pipeline(session, video_id, run_id)
+    logger.info("job done pipeline video_id=%s run_id=%s", video_id, run_id)
+
+
+@dramatiq.actor(max_retries=1)
+def transcribe_job(video_id: int) -> None:
+    logger.info("job start transcribe video_id=%s", video_id)
+    with session_scope() as session:
+        video = session.get(Video, video_id)
+        if video is None:
+            raise RuntimeError(f"Video {video_id} not found.")
+        transcribe_video(session, video)
+    logger.info("job done transcribe video_id=%s", video_id)
+
+
+@dramatiq.actor(max_retries=1)
+def suggest_job(video_id: int) -> None:
+    logger.info("job start suggest video_id=%s", video_id)
+    with session_scope() as session:
+        video = session.get(Video, video_id)
+        if video is None:
+            raise RuntimeError(f"Video {video_id} not found.")
+        suggest_clips(session, video)
+    logger.info("job done suggest video_id=%s", video_id)
+
+
+@dramatiq.actor(max_retries=1)
+def render_job(clip_id: int) -> None:
+    logger.info("job start render clip_id=%s", clip_id)
+    with session_scope() as session:
+        render_clip_by_id(session, clip_id)
+    logger.info("job done render clip_id=%s", clip_id)
+
+
+@dramatiq.actor(max_retries=1)
+def upload_job(clip_id: int) -> None:
+    logger.info("job start upload clip_id=%s", clip_id)
+    with session_scope() as session:
+        upload_clip_by_id(session, clip_id)
+    logger.info("job done upload clip_id=%s", clip_id)
@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+import logging
+import sys
+
+from evanescere.config import get_settings
+
+
+def configure_logging() -> None:
+    settings = get_settings()
+    level = getattr(logging, settings.log_level.upper(), logging.INFO)
+    logging.basicConfig(
+        level=level,
+        format="%(asctime)s %(levelname)s [%(name)s] %(message)s",
+        datefmt="%Y-%m-%dT%H:%M:%S%z",
+        stream=sys.stdout,
+        force=True,
+    )
+    logging.getLogger("httpx").setLevel(logging.DEBUG if level <= logging.DEBUG else logging.WARNING)
+    logging.getLogger("httpcore").setLevel(logging.INFO if level <= logging.DEBUG else logging.WARNING)
+    logging.getLogger("sqlalchemy.engine").setLevel(logging.INFO if settings.log_sql else logging.WARNING)
+    logging.getLogger("evanescere").debug(
+        "logging configured level=%s sql=%s env=%s",
+        settings.log_level,
+        settings.log_sql,
+        settings.app_env,
+    )
@@ -0,0 +1,113 @@
+from datetime import UTC, datetime
+from typing import Any
+
+from sqlalchemy import BigInteger, Boolean, DateTime, Float, ForeignKey, Integer, String, Text
+from sqlalchemy.dialects.postgresql import JSONB
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column, relationship
+
+
+def utcnow() -> datetime:
+    return datetime.now(UTC)
+
+
+class Base(DeclarativeBase):
+    pass
+
+
+class TimestampMixin:
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utcnow)
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), default=utcnow, onupdate=utcnow
+    )
+
+
+class Video(Base, TimestampMixin):
+    __tablename__ = "videos"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True)
+    source_url: Mapped[str] = mapped_column(Text, unique=True, nullable=False)
+    filename: Mapped[str] = mapped_column(Text, nullable=False)
+    size_bytes: Mapped[int | None] = mapped_column(BigInteger)
+    size_samples: Mapped[list[dict[str, Any]]] = mapped_column(JSONB, default=list)
+    duration_sec: Mapped[float | None] = mapped_column(Float)
+    codec_metadata: Mapped[dict[str, Any]] = mapped_column(JSONB, default=dict)
+    ingest_status: Mapped[str] = mapped_column(String(32), default="observing", index=True)
+    processing_status: Mapped[str] = mapped_column(String(32), default="pending")
+
+    runs: Mapped[list["PipelineRun"]] = relationship(back_populates="video")
+    transcript_segments: Mapped[list["TranscriptSegment"]] = relationship(back_populates="video")
+    clip_suggestions: Mapped[list["ClipSuggestion"]] = relationship(back_populates="video")
+
+
+class PipelineRun(Base, TimestampMixin):
+    __tablename__ = "pipeline_runs"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True)
+    video_id: Mapped[int] = mapped_column(ForeignKey("videos.id", ondelete="CASCADE"))
+    trigger: Mapped[str] = mapped_column(String(32), default="manual")
+    stage: Mapped[str] = mapped_column(String(64), default="created")
+    status: Mapped[str] = mapped_column(String(32), default="queued")
+    error: Mapped[str | None] = mapped_column(Text)
+
+    video: Mapped[Video] = relationship(back_populates="runs")
+
+
+class Artifact(Base):
+    __tablename__ = "artifacts"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True)
+    video_id: Mapped[int | None] = mapped_column(ForeignKey("videos.id", ondelete="CASCADE"))
+    clip_id: Mapped[int | None] = mapped_column(Integer)
+    artifact_type: Mapped[str] = mapped_column(String(64), nullable=False)
+    local_path: Mapped[str] = mapped_column(Text, nullable=False)
+    webdav_url: Mapped[str | None] = mapped_column(Text)
+    preserve: Mapped[bool] = mapped_column(Boolean, default=True)
+    artifact_metadata: Mapped[dict[str, Any]] = mapped_column("metadata", JSONB, default=dict)
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utcnow)
+
+
+class TranscriptSegment(Base):
+    __tablename__ = "transcript_segments"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True)
+    video_id: Mapped[int] = mapped_column(ForeignKey("videos.id", ondelete="CASCADE"), index=True)
+    start_sec: Mapped[float] = mapped_column(Float, nullable=False)
+    end_sec: Mapped[float] = mapped_column(Float, nullable=False)
+    text: Mapped[str] = mapped_column(Text, nullable=False)
+    speaker: Mapped[str | None] = mapped_column(Text)
+    confidence: Mapped[float | None] = mapped_column(Float)
+    segment_metadata: Mapped[dict[str, Any]] = mapped_column("metadata", JSONB, default=dict)
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utcnow)
+
+    video: Mapped[Video] = relationship(back_populates="transcript_segments")
+
+
+class ClipSuggestion(Base, TimestampMixin):
+    __tablename__ = "clip_suggestions"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True)
+    video_id: Mapped[int] = mapped_column(ForeignKey("videos.id", ondelete="CASCADE"))
+    start_sec: Mapped[float] = mapped_column(Float, nullable=False)
+    end_sec: Mapped[float] = mapped_column(Float, nullable=False)
+    title_zh: Mapped[str] = mapped_column(Text, nullable=False)
+    summary_zh: Mapped[str] = mapped_column(Text, nullable=False)
+    reason: Mapped[str] = mapped_column(Text, nullable=False)
+    score: Mapped[float] = mapped_column(Float, nullable=False)
+    tags: Mapped[list[str]] = mapped_column(JSONB, default=list)
+    subtitle_priority: Mapped[str] = mapped_column(String(32), default="normal")
+    llm_raw: Mapped[dict[str, Any]] = mapped_column(JSONB, default=dict)
+    approval_status: Mapped[str] = mapped_column(String(32), default="pending")
+    render_status: Mapped[str] = mapped_column(String(32), default="pending")
+    upload_status: Mapped[str] = mapped_column(String(32), default="pending")
+
+    video: Mapped[Video] = relationship(back_populates="clip_suggestions")
+
+
+class Setting(Base):
+    __tablename__ = "settings"
+
+    key: Mapped[str] = mapped_column(String(128), primary_key=True)
+    value: Mapped[dict[str, Any]] = mapped_column(JSONB, nullable=False)
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), default=utcnow, onupdate=utcnow
+    )
@@ -0,0 +1,251 @@
+from __future__ import annotations
+
+import json
+import logging
+from pathlib import Path
+
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import Artifact, ClipSuggestion, PipelineRun, TranscriptSegment, Video
+from evanescere.services.artifacts import latest_artifact, register_artifact, video_work_dir
+from evanescere.services.asr import FunAsrClient, normalize_segments
+from evanescere.services.llm import DeepSeekClipClient, chunk_transcript
+from evanescere.services.media import extract_audio, remux_to_mp4, render_clip
+from evanescere.services.subtitles import generate_clip_subtitles
+from evanescere.services.thumbnail import generate_thumbnail
+from evanescere.services.uploader import choose_upload_artifact, upload_artifact
+from evanescere.services.webdav import WebDavClient
+from evanescere.settings_store import get_pipeline_settings
+
+logger = logging.getLogger(__name__)
+
+
+def update_run(run: PipelineRun | None, stage: str, status: str, error: str | None = None) -> None:
+    if run is not None:
+        logger.info("run update run_id=%s video_id=%s stage=%s status=%s", run.id, run.video_id, stage, status)
+        run.stage = stage
+        run.status = status
+        run.error = error
+
+
+def source_artifact(session: Session, video: Video) -> Artifact | None:
+    return latest_artifact(session, video.id, "raw_source")
+
+
+def download_source(session: Session, video: Video) -> Path:
+    existing = source_artifact(session, video)
+    if existing and Path(existing.local_path).exists():
+        logger.info("source already downloaded video_id=%s path=%s", video.id, existing.local_path)
+        return Path(existing.local_path)
+    destination = video_work_dir(video.id) / video.filename
+    logger.info("downloading source video_id=%s url=%s destination=%s", video.id, video.source_url, destination)
+    WebDavClient.from_settings().download(video.source_url, destination)
+    register_artifact(
+        session,
+        video_id=video.id,
+        artifact_type="raw_source",
+        local_path=destination,
+        preserve=False,
+        metadata={"source_url": video.source_url},
+    )
+    return destination
+
+
+def prepare_media(session: Session, video: Video) -> tuple[Path, Path]:
+    logger.info("prepare media start video_id=%s file=%s", video.id, video.filename)
+    source = download_source(session, video)
+    mp4_artifact = latest_artifact(session, video.id, "remux_mp4")
+    if mp4_artifact and Path(mp4_artifact.local_path).exists():
+        logger.info("using existing remux video_id=%s path=%s", video.id, mp4_artifact.local_path)
+        mp4 = Path(mp4_artifact.local_path)
+    else:
+        mp4 = remux_to_mp4(session, video, source)
+    audio_artifact = latest_artifact(session, video.id, "audio_wav")
+    if audio_artifact and Path(audio_artifact.local_path).exists():
+        logger.info("using existing audio video_id=%s path=%s", video.id, audio_artifact.local_path)
+        audio = Path(audio_artifact.local_path)
+    else:
+        audio = extract_audio(session, video, mp4)
+    logger.info("prepare media done video_id=%s mp4=%s audio=%s", video.id, mp4, audio)
+    return mp4, audio
+
+
+def transcribe_video(session: Session, video: Video) -> int:
+    logger.info("transcription start video_id=%s", video.id)
+    audio_artifact = latest_artifact(session, video.id, "audio_wav")
+    if audio_artifact is None:
+        raise RuntimeError("Cannot transcribe before audio_wav artifact exists.")
+    payload = FunAsrClient.from_settings().transcribe(Path(audio_artifact.local_path))
+    transcript_path = video_work_dir(video.id) / "funasr_transcript.json"
+    transcript_path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+    register_artifact(
+        session,
+        video_id=video.id,
+        artifact_type="transcript_json",
+        local_path=transcript_path,
+        metadata={"source": "funasr"},
+    )
+    session.query(TranscriptSegment).filter(TranscriptSegment.video_id == video.id).delete()
+    for item in normalize_segments(payload):
+        session.add(
+            TranscriptSegment(
+                video_id=video.id,
+                start_sec=item["start_sec"],
+                end_sec=item["end_sec"],
+                text=item["text"],
+                speaker=item["speaker"],
+                confidence=item["confidence"],
+                segment_metadata=item["metadata"],
+            )
+        )
+    session.flush()
+    count = session.query(TranscriptSegment).filter(TranscriptSegment.video_id == video.id).count()
+    logger.info("transcription done video_id=%s segments=%s transcript=%s", video.id, count, transcript_path)
+    return count
+
+
+def suggest_clips(session: Session, video: Video) -> int:
+    settings = get_settings()
+    logger.info("clip suggestion start video_id=%s chunk_seconds=%s", video.id, settings.transcript_chunk_seconds)
+    segments = list(
+        session.scalars(
+            select(TranscriptSegment)
+            .where(TranscriptSegment.video_id == video.id)
+            .order_by(TranscriptSegment.start_sec)
+        ).all()
+    )
+    if not segments:
+        raise RuntimeError("Cannot suggest clips without transcript segments.")
+    chunks = chunk_transcript(segments, settings.transcript_chunk_seconds)
+    logger.info("clip suggestion chunks video_id=%s segments=%s chunks=%s", video.id, len(segments), len(chunks))
+    candidates = DeepSeekClipClient.from_settings().suggest(chunks)
+    session.query(ClipSuggestion).filter(ClipSuggestion.video_id == video.id).delete()
+    for candidate in candidates:
+        session.add(
+            ClipSuggestion(
+                video_id=video.id,
+                start_sec=candidate.start_sec,
+                end_sec=candidate.end_sec,
+                title_zh=candidate.title_zh,
+                summary_zh=candidate.summary_zh,
+                reason=candidate.reason,
+                score=candidate.score,
+                tags=candidate.tags,
+                subtitle_priority=candidate.subtitle_priority,
+                llm_raw=candidate.model_dump(),
+            )
+        )
+    session.flush()
+    logger.info("clip suggestion done video_id=%s candidates=%s", video.id, len(candidates))
+    return len(candidates)
+
+
+def render_clip_by_id(session: Session, clip_id: int) -> None:
+    logger.info("render start clip_id=%s", clip_id)
+    clip = session.get(ClipSuggestion, clip_id)
+    if clip is None:
+        raise RuntimeError(f"Clip {clip_id} not found.")
+    video = session.get(Video, clip.video_id)
+    if video is None:
+        raise RuntimeError(f"Video {clip.video_id} not found.")
+    mp4_artifact = latest_artifact(session, video.id, "remux_mp4")
+    if mp4_artifact is None:
+        raise RuntimeError("Cannot render before remux_mp4 artifact exists.")
+    settings = get_pipeline_settings(session)
+    segments = list(
+        session.scalars(
+            select(TranscriptSegment)
+            .where(TranscriptSegment.video_id == video.id)
+            .order_by(TranscriptSegment.start_sec)
+        ).all()
+    )
+    clip.render_status = "running"
+    srt, ass = generate_clip_subtitles(
+        session,
+        video=video,
+        clip=clip,
+        segments=segments,
+        preserve=settings.preserve_final_artifacts,
+    )
+    render_clip(
+        session,
+        video=video,
+        clip=clip,
+        source_mp4=Path(mp4_artifact.local_path),
+        srt_path=srt,
+        ass_path=ass,
+        bake_subtitles=settings.bake_subtitles,
+        preserve=settings.preserve_final_artifacts,
+    )
+    generate_thumbnail(
+        session,
+        video=video,
+        clip=clip,
+        source_mp4=Path(mp4_artifact.local_path),
+        transcript_segments=segments,
+    )
+    clip.render_status = "done"
+    logger.info("render done clip_id=%s video_id=%s", clip.id, video.id)
+
+
+def upload_clip_by_id(session: Session, clip_id: int) -> None:
+    logger.info("upload start clip_id=%s", clip_id)
+    clip = session.get(ClipSuggestion, clip_id)
+    if clip is None:
+        raise RuntimeError(f"Clip {clip_id} not found.")
+    clip.upload_status = "running"
+    artifacts = (
+        session.query(Artifact)
+        .filter(Artifact.clip_id == clip_id)
+        .order_by(Artifact.created_at.desc())
+        .all()
+    )
+    artifact = choose_upload_artifact(artifacts)
+    if artifact is None:
+        raise RuntimeError("No rendered clip artifact found for upload.")
+    result = upload_artifact(session, clip, artifact)
+    artifact.artifact_metadata = artifact.artifact_metadata | {"upload": dict(result)}
+    clip.upload_status = "done"
+    logger.info("upload done clip_id=%s result=%s", clip.id, dict(result))
+
+
+def run_pipeline(session: Session, video_id: int, run_id: int | None = None) -> None:
+    logger.info("pipeline start video_id=%s run_id=%s", video_id, run_id)
+    video = session.get(Video, video_id)
+    if video is None:
+        raise RuntimeError(f"Video {video_id} not found.")
+    run = session.get(PipelineRun, run_id) if run_id is not None else None
+    settings = get_pipeline_settings(session)
+    try:
+        video.processing_status = "running"
+        update_run(run, "media", "running")
+        prepare_media(session, video)
+
+        update_run(run, "transcribe", "running")
+        transcribe_video(session, video)
+
+        if settings.suggest_enabled:
+            update_run(run, "suggest", "running")
+            suggest_clips(session, video)
+
+        if settings.render_enabled:
+            clips = session.scalars(
+                select(ClipSuggestion).where(ClipSuggestion.video_id == video.id)
+            ).all()
+            logger.info("auto render stage video_id=%s clips=%s upload_enabled=%s", video.id, len(clips), settings.upload_enabled)
+            for clip in clips:
+                clip.approval_status = "auto_approved"
+                render_clip_by_id(session, clip.id)
+                if settings.upload_enabled:
+                    upload_clip_by_id(session, clip.id)
+
+        video.processing_status = "done"
+        update_run(run, "done", "done")
+        logger.info("pipeline done video_id=%s run_id=%s", video_id, run_id)
+    except Exception as exc:
+        video.processing_status = "failed"
+        update_run(run, "failed", "failed", str(exc))
+        logger.exception("pipeline failed video_id=%s run_id=%s", video_id, run_id)
+        raise
@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+import logging
+import time
+
+from sqlalchemy import select
+
+from evanescere.config import get_settings
+from evanescere.db import session_scope
+from evanescere.jobs import enqueue_pipeline
+from evanescere.logging_config import configure_logging
+from evanescere.models import PipelineRun, Video
+from evanescere.services.webdav import WebDavClient, scan_once
+
+configure_logging()
+logger = logging.getLogger(__name__)
+
+
+def enqueue_stable_videos() -> int:
+    queued = 0
+    with session_scope() as session:
+        videos = session.scalars(
+            select(Video).where(Video.ingest_status == "stable", Video.processing_status == "pending")
+        ).all()
+        for video in videos:
+            run = PipelineRun(video_id=video.id, trigger="automatic", stage="queued", status="queued")
+            video.ingest_status = "queued"
+            video.processing_status = "queued"
+            session.add(run)
+            session.flush()
+            logger.info("queueing stable video video_id=%s run_id=%s file=%s", video.id, run.id, video.filename)
+            enqueue_pipeline(video.id, run.id)
+            queued += 1
+    return queued
+
+
+def run_forever() -> None:
+    settings = get_settings()
+    client = WebDavClient.from_settings()
+    logger.info(
+        "scheduler starting webdav_base=%s poll_interval_seconds=%s",
+        settings.webdav_base_url,
+        settings.webdav_poll_interval_seconds,
+    )
+    while True:
+        try:
+            with session_scope() as session:
+                observed, stable = scan_once(client, session)
+            queued = enqueue_stable_videos()
+            logger.info("scan observed=%s newly_stable=%s queued=%s", observed, stable, queued)
+        except Exception:
+            logger.exception("scheduler scan failed")
+        time.sleep(settings.webdav_poll_interval_seconds)
+
+
+if __name__ == "__main__":
+    run_forever()
@@ -0,0 +1,111 @@
+from datetime import datetime
+from typing import Any
+
+from pydantic import BaseModel, ConfigDict, Field
+
+
+class VideoRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: int
+    source_url: str
+    filename: str
+    size_bytes: int | None
+    duration_sec: float | None
+    codec_metadata: dict[str, Any]
+    ingest_status: str
+    processing_status: str
+    created_at: datetime
+    updated_at: datetime
+
+
+class RunRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: int
+    video_id: int
+    trigger: str
+    stage: str
+    status: str
+    error: str | None
+    created_at: datetime
+    updated_at: datetime
+
+
+class ArtifactRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: int
+    video_id: int | None
+    clip_id: int | None
+    artifact_type: str
+    local_path: str
+    webdav_url: str | None
+    preserve: bool
+    artifact_metadata: dict[str, Any]
+    created_at: datetime
+
+
+class TranscriptSegmentRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: int
+    video_id: int
+    start_sec: float
+    end_sec: float
+    text: str
+    speaker: str | None
+    confidence: float | None
+    segment_metadata: dict[str, Any]
+
+
+class ClipSuggestionRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: int
+    video_id: int
+    start_sec: float
+    end_sec: float
+    title_zh: str
+    summary_zh: str
+    reason: str
+    score: float
+    tags: list[str]
+    subtitle_priority: str
+    approval_status: str
+    render_status: str
+    upload_status: str
+    created_at: datetime
+    updated_at: datetime
+
+
+class SettingsPatch(BaseModel):
+    suggest_enabled: bool | None = None
+    render_enabled: bool | None = None
+    upload_enabled: bool | None = None
+    preserve_final_artifacts: bool | None = None
+    bake_subtitles: bool | None = None
+
+
+class SettingsRead(BaseModel):
+    suggest_enabled: bool = True
+    render_enabled: bool = True
+    upload_enabled: bool = True
+    preserve_final_artifacts: bool = True
+    bake_subtitles: bool = True
+
+
+class ClipCandidate(BaseModel):
+    start_sec: float = Field(ge=0)
+    end_sec: float = Field(gt=0)
+    title_zh: str
+    summary_zh: str
+    reason: str
+    score: float = Field(ge=0, le=1)
+    tags: list[str] = Field(default_factory=list)
+    subtitle_priority: str = "normal"
+
+
+class ClipCandidateResponse(BaseModel):
+    clips: list[ClipCandidate]
+
@@ -0,0 +1,2 @@
+"""Service adapters for storage, AI, media, subtitles, and upload."""
+
@@ -0,0 +1,60 @@
+from pathlib import Path
+
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import Artifact
+
+
+def video_work_dir(video_id: int) -> Path:
+    root = get_settings().local_storage_root / "videos" / str(video_id)
+    root.mkdir(parents=True, exist_ok=True)
+    return root
+
+
+def clip_work_dir(video_id: int, clip_id: int) -> Path:
+    root = video_work_dir(video_id) / "clips" / str(clip_id)
+    root.mkdir(parents=True, exist_ok=True)
+    return root
+
+
+def register_artifact(
+    session: Session,
+    *,
+    artifact_type: str,
+    local_path: Path,
+    video_id: int | None = None,
+    clip_id: int | None = None,
+    preserve: bool = True,
+    metadata: dict | None = None,
+) -> Artifact:
+    artifact = Artifact(
+        video_id=video_id,
+        clip_id=clip_id,
+        artifact_type=artifact_type,
+        local_path=str(local_path),
+        preserve=preserve,
+        artifact_metadata=metadata or {},
+    )
+    session.add(artifact)
+    session.flush()
+    return artifact
+
+
+def latest_artifact(session: Session, video_id: int, artifact_type: str) -> Artifact | None:
+    return (
+        session.query(Artifact)
+        .filter(Artifact.video_id == video_id, Artifact.artifact_type == artifact_type)
+        .order_by(Artifact.created_at.desc())
+        .first()
+    )
+
+
+def latest_clip_artifact(session: Session, clip_id: int, artifact_type: str) -> Artifact | None:
+    return (
+        session.query(Artifact)
+        .filter(Artifact.clip_id == clip_id, Artifact.artifact_type == artifact_type)
+        .order_by(Artifact.created_at.desc())
+        .first()
+    )
+
@@ -0,0 +1,85 @@
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Any
+
+import httpx
+
+from evanescere.config import get_settings
+
+logger = logging.getLogger(__name__)
+
+
+class FunAsrClient:
+    def __init__(self, base_url: str, api_key: str | None, model: str) -> None:
+        self.base_url = base_url.rstrip("/")
+        self.api_key = api_key
+        self.model = model
+
+    @classmethod
+    def from_settings(cls) -> FunAsrClient:
+        settings = get_settings()
+        return cls(settings.funasr_base_url, settings.funasr_api_key, settings.funasr_model)
+
+    def transcribe(self, audio_path: Path) -> dict[str, Any]:
+        logger.info("funasr request start base_url=%s model=%s audio=%s", self.base_url, self.model, audio_path)
+        headers = {}
+        if self.api_key:
+            headers["Authorization"] = f"Bearer {self.api_key}"
+        with open(audio_path, "rb") as audio_file:
+            response = httpx.post(
+                f"{self.base_url}/audio/transcriptions",
+                headers=headers,
+                data={
+                    "model": self.model,
+                    "language": "zh",
+                    "response_format": "verbose_json",
+                    "timestamp_granularities[]": "segment",
+                },
+                files={"file": (audio_path.name, audio_file, "audio/wav")},
+                timeout=None,
+            )
+        response.raise_for_status()
+        payload = response.json()
+        logger.info("funasr request done audio=%s keys=%s", audio_path, sorted(payload.keys()))
+        return payload
+
+
+def normalize_segments(payload: dict[str, Any]) -> list[dict[str, Any]]:
+    raw_segments = payload.get("segments") or payload.get("sentence_info") or []
+    logger.debug("normalizing asr segments raw_count=%s", len(raw_segments))
+    normalized: list[dict[str, Any]] = []
+    for item in raw_segments:
+        start = item.get("start") or item.get("start_sec") or item.get("timestamp", [0, 0])[0]
+        end = item.get("end") or item.get("end_sec") or item.get("timestamp", [start, start])[1]
+        if isinstance(start, int) and start > 10_000:
+            start = start / 1000
+        if isinstance(end, int) and end > 10_000:
+            end = end / 1000
+        text = item.get("text") or item.get("sentence") or item.get("raw_text") or ""
+        if not text.strip():
+            continue
+        normalized.append(
+            {
+                "start_sec": float(start),
+                "end_sec": float(end),
+                "text": text.strip(),
+                "speaker": item.get("speaker"),
+                "confidence": item.get("confidence"),
+                "metadata": item,
+            }
+        )
+    if not normalized and payload.get("text"):
+        normalized.append(
+            {
+                "start_sec": 0.0,
+                "end_sec": 0.1,
+                "text": str(payload["text"]).strip(),
+                "speaker": None,
+                "confidence": None,
+                "metadata": payload,
+            }
+        )
+    logger.info("normalized asr segments count=%s", len(normalized))
+    return normalized
@@ -0,0 +1,157 @@
+from __future__ import annotations
+
+import json
+import logging
+from dataclasses import dataclass
+
+from openai import OpenAI
+from pydantic import ValidationError
+
+from evanescere.config import get_settings
+from evanescere.models import TranscriptSegment
+from evanescere.schemas import ClipCandidate, ClipCandidateResponse
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass(frozen=True)
+class TranscriptChunk:
+    start_sec: float
+    end_sec: float
+    text: str
+
+
+class DeepSeekClipClient:
+    def __init__(self, api_key: str, base_url: str, model: str, temperature: float) -> None:
+        self.client = OpenAI(api_key=api_key, base_url=base_url)
+        self.model = model
+        self.temperature = temperature
+
+    @classmethod
+    def from_settings(cls) -> DeepSeekClipClient:
+        settings = get_settings()
+        if not settings.deepseek_api_key:
+            raise RuntimeError("[deepseek].api_key is required for clip suggestion.")
+        return cls(
+            settings.deepseek_api_key,
+            settings.deepseek_base_url,
+            settings.deepseek_model,
+            settings.deepseek_temperature,
+        )
+
+    def suggest(self, chunks: list[TranscriptChunk]) -> list[ClipCandidate]:
+        logger.info("deepseek suggestion start chunks=%s model=%s", len(chunks), self.model)
+        candidates: list[ClipCandidate] = []
+        for chunk in chunks:
+            candidates.extend(self._suggest_for_chunk(chunk))
+        ranked = dedupe_and_rank_candidates(candidates)
+        logger.info("deepseek suggestion done raw_candidates=%s ranked_candidates=%s", len(candidates), len(ranked))
+        return ranked
+
+    def _suggest_for_chunk(self, chunk: TranscriptChunk) -> list[ClipCandidate]:
+        logger.info(
+            "deepseek chunk request start start=%.1f end=%.1f chars=%s",
+            chunk.start_sec,
+            chunk.end_sec,
+            len(chunk.text),
+        )
+        response = self.client.chat.completions.create(
+            model=self.model,
+            temperature=self.temperature,
+            response_format={"type": "json_object"},
+            messages=[
+                {
+                    "role": "system",
+                    "content": (
+                        "You select interesting short clips from Mandarin livestream transcripts. "
+                        "Return strict JSON only. Prefer complete moments with context, funny reactions, "
+                        "surprising reveals, emotional peaks, or strong standalone discussion."
+                    ),
+                },
+                {
+                    "role": "user",
+                    "content": (
+                        "Find up to 5 clip candidates in this transcript chunk. "
+                        "Each clip must be 30-360 seconds and use absolute stream seconds. "
+                        "JSON schema: {\"clips\":[{\"start_sec\":number,\"end_sec\":number,"
+                        "\"title_zh\":string,\"summary_zh\":string,\"reason\":string,"
+                        "\"score\":number,\"tags\":[string],\"subtitle_priority\":string}]}.\n\n"
+                        f"Chunk bounds: {chunk.start_sec:.1f}-{chunk.end_sec:.1f}\n"
+                        f"Transcript:\n{chunk.text}"
+                    ),
+                },
+            ],
+        )
+        content = response.choices[0].message.content or "{}"
+        candidates = parse_clip_response(content)
+        usage = getattr(response, "usage", None)
+        logger.info(
+            "deepseek chunk request done start=%.1f end=%.1f candidates=%s usage=%s",
+            chunk.start_sec,
+            chunk.end_sec,
+            len(candidates),
+            usage,
+        )
+        return candidates
+
+
+def segment_line(segment: TranscriptSegment) -> str:
+    return f"[{segment.start_sec:.1f}-{segment.end_sec:.1f}] {segment.text}"
+
+
+def chunk_transcript(
+    segments: list[TranscriptSegment], chunk_seconds: int
+) -> list[TranscriptChunk]:
+    chunks: list[TranscriptChunk] = []
+    current: list[str] = []
+    current_start: float | None = None
+    current_end: float | None = None
+    for segment in segments:
+        if current_start is None:
+            current_start = segment.start_sec
+        if current_end is not None and segment.end_sec - current_start > chunk_seconds:
+            chunks.append(
+                TranscriptChunk(current_start, current_end, "\n".join(current))
+            )
+            current = []
+            current_start = segment.start_sec
+        current.append(segment_line(segment))
+        current_end = segment.end_sec
+    if current and current_start is not None and current_end is not None:
+        chunks.append(TranscriptChunk(current_start, current_end, "\n".join(current)))
+    logger.debug("chunked transcript segments=%s chunk_seconds=%s chunks=%s", len(segments), chunk_seconds, len(chunks))
+    return chunks
+
+
+def parse_clip_response(content: str) -> list[ClipCandidate]:
+    try:
+        payload = json.loads(content)
+        parsed = ClipCandidateResponse.model_validate(payload)
+    except (json.JSONDecodeError, ValidationError) as exc:
+        logger.error("invalid clip json content_prefix=%s", content[:1000])
+        raise ValueError(f"Invalid clip JSON from LLM: {exc}") from exc
+    return parsed.clips
+
+
+def overlap_ratio(left: ClipCandidate, right: ClipCandidate) -> float:
+    overlap = max(0, min(left.end_sec, right.end_sec) - max(left.start_sec, right.start_sec))
+    shortest = min(left.end_sec - left.start_sec, right.end_sec - right.start_sec)
+    if shortest <= 0:
+        return 0
+    return overlap / shortest
+
+
+def dedupe_and_rank_candidates(candidates: list[ClipCandidate]) -> list[ClipCandidate]:
+    valid = [
+        candidate
+        for candidate in candidates
+        if candidate.end_sec > candidate.start_sec
+        and 30 <= candidate.end_sec - candidate.start_sec <= 360
+    ]
+    ranked = sorted(valid, key=lambda item: item.score, reverse=True)
+    chosen: list[ClipCandidate] = []
+    for candidate in ranked:
+        if all(overlap_ratio(candidate, existing) < 0.5 for existing in chosen):
+            chosen.append(candidate)
+    logger.debug("dedupe candidates input=%s valid=%s chosen=%s", len(candidates), len(valid), len(chosen))
+    return chosen
@@ -0,0 +1,201 @@
+from __future__ import annotations
+
+import json
+import logging
+import subprocess
+from pathlib import Path
+
+from sqlalchemy.orm import Session
+
+from evanescere.models import ClipSuggestion, Video
+from evanescere.services.artifacts import clip_work_dir, register_artifact, video_work_dir
+
+logger = logging.getLogger(__name__)
+
+
+def run_command(args: list[str]) -> None:
+    logger.debug("command start args=%s", args)
+    completed = subprocess.run(args, check=False, text=True, capture_output=True)
+    if completed.returncode != 0:
+        logger.error("command failed returncode=%s stderr=%s", completed.returncode, completed.stderr[-4000:])
+        raise RuntimeError(
+            f"Command failed ({completed.returncode}): {' '.join(args)}\n{completed.stderr}"
+        )
+    logger.debug("command done args=%s stderr_tail=%s", args, completed.stderr[-1000:])
+
+
+def ffprobe(path: Path) -> dict:
+    logger.debug("ffprobe start path=%s", path)
+    completed = subprocess.run(
+        [
+            "ffprobe",
+            "-v",
+            "error",
+            "-print_format",
+            "json",
+            "-show_format",
+            "-show_streams",
+            str(path),
+        ],
+        check=True,
+        text=True,
+        capture_output=True,
+    )
+    metadata = json.loads(completed.stdout)
+    logger.debug("ffprobe done path=%s streams=%s", path, len(metadata.get("streams", [])))
+    return metadata
+
+
+def duration_from_probe(metadata: dict) -> float | None:
+    duration = metadata.get("format", {}).get("duration")
+    return float(duration) if duration else None
+
+
+def remux_to_mp4(session: Session, video: Video, source_path: Path) -> Path:
+    output = video_work_dir(video.id) / f"{Path(video.filename).stem}.mp4"
+    logger.info("remux start video_id=%s source=%s output=%s", video.id, source_path, output)
+    try:
+        run_command(["ffmpeg", "-y", "-i", str(source_path), "-map", "0", "-c", "copy", str(output)])
+    except RuntimeError:
+        logger.warning("remux copy failed, falling back to transcode video_id=%s", video.id)
+        run_command(
+            [
+                "ffmpeg",
+                "-y",
+                "-i",
+                str(source_path),
+                "-c:v",
+                "libx264",
+                "-preset",
+                "veryfast",
+                "-crf",
+                "20",
+                "-c:a",
+                "aac",
+                str(output),
+            ]
+        )
+    metadata = ffprobe(output)
+    video.duration_sec = duration_from_probe(metadata)
+    video.codec_metadata = metadata
+    register_artifact(session, video_id=video.id, artifact_type="remux_mp4", local_path=output)
+    logger.info("remux done video_id=%s duration=%s output=%s", video.id, video.duration_sec, output)
+    return output
+
+
+def extract_audio(session: Session, video: Video, media_path: Path) -> Path:
+    output = video_work_dir(video.id) / "audio_16k_mono.wav"
+    logger.info("audio extraction start video_id=%s source=%s output=%s", video.id, media_path, output)
+    run_command(
+        [
+            "ffmpeg",
+            "-y",
+            "-i",
+            str(media_path),
+            "-vn",
+            "-ac",
+            "1",
+            "-ar",
+            "16000",
+            "-c:a",
+            "pcm_s16le",
+            str(output),
+        ]
+    )
+    register_artifact(session, video_id=video.id, artifact_type="audio_wav", local_path=output)
+    logger.info("audio extraction done video_id=%s output=%s", video.id, output)
+    return output
+
+
+def render_clip(
+    session: Session,
+    *,
+    video: Video,
+    clip: ClipSuggestion,
+    source_mp4: Path,
+    srt_path: Path,
+    ass_path: Path,
+    bake_subtitles: bool,
+    preserve: bool,
+) -> list[Path]:
+    output_dir = clip_work_dir(video.id, clip.id)
+    logger.info(
+        "render clip media start video_id=%s clip_id=%s start=%.3f end=%.3f bake=%s",
+        video.id,
+        clip.id,
+        clip.start_sec,
+        clip.end_sec,
+        bake_subtitles,
+    )
+    base_args = [
+        "ffmpeg",
+        "-y",
+        "-ss",
+        f"{clip.start_sec:.3f}",
+        "-to",
+        f"{clip.end_sec:.3f}",
+        "-i",
+        str(source_mp4),
+    ]
+    rendered: list[Path] = []
+
+    soft_path = output_dir / "clip_soft_sub.mp4"
+    run_command(base_args + ["-c", "copy", str(soft_path)])
+    register_artifact(
+        session,
+        video_id=video.id,
+        clip_id=clip.id,
+        artifact_type="clip_soft_sub_mp4",
+        local_path=soft_path,
+        preserve=preserve,
+        metadata={"srt": str(srt_path), "ass": str(ass_path)},
+    )
+    rendered.append(soft_path)
+
+    if bake_subtitles:
+        baked_path = output_dir / "clip_baked_sub.mp4"
+        run_command(
+            base_args
+            + [
+                "-vf",
+                f"ass={ass_path}",
+                "-c:v",
+                "libx264",
+                "-preset",
+                "veryfast",
+                "-crf",
+                "20",
+                "-c:a",
+                "aac",
+                str(baked_path),
+            ]
+        )
+        register_artifact(
+            session,
+            video_id=video.id,
+            clip_id=clip.id,
+            artifact_type="clip_baked_sub_mp4",
+            local_path=baked_path,
+            preserve=preserve,
+            metadata={"srt": str(srt_path), "ass": str(ass_path)},
+        )
+        rendered.append(baked_path)
+
+    logger.info("render clip media done video_id=%s clip_id=%s outputs=%s", video.id, clip.id, rendered)
+    return rendered
+
+
+def extract_thumbnail_frame(session: Session, video: Video, clip: ClipSuggestion, source_mp4: Path) -> Path:
+    output = clip_work_dir(video.id, clip.id) / "thumbnail_base.jpg"
+    timestamp = max(clip.start_sec, (clip.start_sec + clip.end_sec) / 2)
+    logger.info("thumbnail extraction start video_id=%s clip_id=%s timestamp=%.3f", video.id, clip.id, timestamp)
+    run_command(["ffmpeg", "-y", "-ss", f"{timestamp:.3f}", "-i", str(source_mp4), "-frames:v", "1", str(output)])
+    register_artifact(
+        session,
+        video_id=video.id,
+        clip_id=clip.id,
+        artifact_type="thumbnail_base",
+        local_path=output,
+    )
+    logger.info("thumbnail extraction done video_id=%s clip_id=%s output=%s", video.id, clip.id, output)
+    return output
@@ -0,0 +1,109 @@
+from pathlib import Path
+import logging
+
+from sqlalchemy.orm import Session
+
+from evanescere.models import ClipSuggestion, TranscriptSegment, Video
+from evanescere.services.artifacts import clip_work_dir, register_artifact
+
+logger = logging.getLogger(__name__)
+
+
+def format_srt_time(seconds: float) -> str:
+    millis = round(seconds * 1000)
+    hours, remainder = divmod(millis, 3_600_000)
+    minutes, remainder = divmod(remainder, 60_000)
+    secs, millis = divmod(remainder, 1000)
+    return f"{hours:02}:{minutes:02}:{secs:02},{millis:03}"
+
+
+def format_ass_time(seconds: float) -> str:
+    centis = round(seconds * 100)
+    hours, remainder = divmod(centis, 360_000)
+    minutes, remainder = divmod(remainder, 6_000)
+    secs, centis = divmod(remainder, 100)
+    return f"{hours}:{minutes:02}:{secs:02}.{centis:02}"
+
+
+def clip_segments(segments: list[TranscriptSegment], clip: ClipSuggestion) -> list[TranscriptSegment]:
+    return [segment for segment in segments if segment.end_sec > clip.start_sec and segment.start_sec < clip.end_sec]
+
+
+def write_srt(path: Path, segments: list[TranscriptSegment], offset_sec: float = 0) -> None:
+    lines: list[str] = []
+    for index, segment in enumerate(segments, start=1):
+        start = max(0, segment.start_sec - offset_sec)
+        end = max(start + 0.1, segment.end_sec - offset_sec)
+        lines.extend(
+            [
+                str(index),
+                f"{format_srt_time(start)} --> {format_srt_time(end)}",
+                segment.text.strip(),
+                "",
+            ]
+        )
+    path.write_text("\n".join(lines), encoding="utf-8")
+
+
+def write_ass(path: Path, segments: list[TranscriptSegment], offset_sec: float = 0) -> None:
+    header = """[Script Info]
+ScriptType: v4.00+
+WrapStyle: 0
+ScaledBorderAndShadow: yes
+
+[V4+ Styles]
+Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
+Style: Default,Noto Sans CJK SC,48,&H00FFFFFF,&H000000FF,&H00000000,&H80000000,0,0,0,0,100,100,0,0,1,3,1,2,80,80,60,1
+
+[Events]
+Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
+"""
+    events = []
+    for segment in segments:
+        start = max(0, segment.start_sec - offset_sec)
+        end = max(start + 0.1, segment.end_sec - offset_sec)
+        text = segment.text.replace("\n", r"\N").replace(",", "，")
+        events.append(
+            f"Dialogue: 0,{format_ass_time(start)},{format_ass_time(end)},Default,,0,0,0,,{text}"
+        )
+    path.write_text(header + "\n".join(events) + "\n", encoding="utf-8")
+
+
+def generate_clip_subtitles(
+    session: Session,
+    *,
+    video: Video,
+    clip: ClipSuggestion,
+    segments: list[TranscriptSegment],
+    preserve: bool,
+) -> tuple[Path, Path]:
+    selected = clip_segments(segments, clip)
+    logger.info(
+        "subtitle generation start video_id=%s clip_id=%s selected_segments=%s",
+        video.id,
+        clip.id,
+        len(selected),
+    )
+    output_dir = clip_work_dir(video.id, clip.id)
+    srt = output_dir / "clip.srt"
+    ass = output_dir / "clip.ass"
+    write_srt(srt, selected, offset_sec=clip.start_sec)
+    write_ass(ass, selected, offset_sec=clip.start_sec)
+    register_artifact(
+        session,
+        video_id=video.id,
+        clip_id=clip.id,
+        artifact_type="subtitle_srt",
+        local_path=srt,
+        preserve=preserve,
+    )
+    register_artifact(
+        session,
+        video_id=video.id,
+        clip_id=clip.id,
+        artifact_type="subtitle_ass",
+        local_path=ass,
+        preserve=preserve,
+    )
+    logger.info("subtitle generation done video_id=%s clip_id=%s srt=%s ass=%s", video.id, clip.id, srt, ass)
+    return srt, ass
@@ -0,0 +1,270 @@
+from __future__ import annotations
+
+import json
+import logging
+import shlex
+import subprocess
+from pathlib import Path
+
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import ClipSuggestion, TranscriptSegment, Video
+from evanescere.services.artifacts import clip_work_dir, register_artifact
+from evanescere.services.media import extract_thumbnail_frame
+
+logger = logging.getLogger(__name__)
+
+FONT_CANDIDATES = [
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
+    "/usr/share/fonts/truetype/noto/NotoSansCJK-Regular.ttc",
+    "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
+]
+
+
+def generate_thumbnail(
+    session: Session,
+    *,
+    video: Video,
+    clip: ClipSuggestion,
+    source_mp4: Path,
+    transcript_segments: list[TranscriptSegment],
+) -> Path | None:
+    settings = get_settings()
+    if not settings.thumbnail_enabled:
+        logger.info("thumbnail disabled video_id=%s clip_id=%s", video.id, clip.id)
+        return None
+
+    output_dir = clip_work_dir(video.id, clip.id)
+    base_frame = extract_thumbnail_frame(session, video, clip, source_mp4)
+    provider = settings.thumbnail_provider.lower()
+    logger.info("thumbnail generation start video_id=%s clip_id=%s provider=%s", video.id, clip.id, provider)
+
+    if provider == "frame_overlay":
+        background = base_frame
+    elif provider == "command":
+        background = run_thumbnail_command(
+            output_dir=output_dir,
+            source_frame=base_frame,
+            video=video,
+            clip=clip,
+            transcript_segments=transcript_segments,
+        )
+        register_artifact(
+            session,
+            video_id=video.id,
+            clip_id=clip.id,
+            artifact_type="thumbnail_generated",
+            local_path=background,
+            metadata={"provider": "command"},
+        )
+    else:
+        raise RuntimeError(f"Unsupported [thumbnail].provider={settings.thumbnail_provider!r}")
+
+    final_path = output_dir / "thumbnail_final.jpg"
+    character_path = (
+        Path(settings.thumbnail_character_overlay_path)
+        if settings.thumbnail_character_overlay_path
+        else None
+    )
+    compose_thumbnail(
+        background_path=background,
+        output_path=final_path,
+        title=clip.title_zh,
+        character_overlay_path=character_path,
+        character_scale=settings.thumbnail_character_scale,
+        character_position=settings.thumbnail_character_position,
+        title_enabled=settings.thumbnail_title_enabled,
+        width=settings.thumbnail_width,
+        height=settings.thumbnail_height,
+    )
+    register_artifact(
+        session,
+        video_id=video.id,
+        clip_id=clip.id,
+        artifact_type="thumbnail_final",
+        local_path=final_path,
+        preserve=True,
+        metadata={
+            "provider": provider,
+            "source_frame": str(base_frame),
+            "background": str(background),
+            "character_overlay_path": settings.thumbnail_character_overlay_path or "",
+        },
+    )
+    logger.info("thumbnail generation done video_id=%s clip_id=%s output=%s", video.id, clip.id, final_path)
+    return final_path
+
+
+def run_thumbnail_command(
+    *,
+    output_dir: Path,
+    source_frame: Path,
+    video: Video,
+    clip: ClipSuggestion,
+    transcript_segments: list[TranscriptSegment],
+) -> Path:
+    settings = get_settings()
+    if not settings.thumbnail_command:
+        raise RuntimeError('[thumbnail].command must be set when [thumbnail].provider = "command"')
+
+    output_path = output_dir / "thumbnail_generated.png"
+    nearby_transcript = [
+        {
+            "start_sec": segment.start_sec,
+            "end_sec": segment.end_sec,
+            "text": segment.text,
+        }
+        for segment in transcript_segments
+        if segment.end_sec > clip.start_sec and segment.start_sec < clip.end_sec
+    ][:80]
+    payload = {
+        "video_id": video.id,
+        "clip_id": clip.id,
+        "source_frame": str(source_frame),
+        "output_path": str(output_path),
+        "width": settings.thumbnail_width,
+        "height": settings.thumbnail_height,
+        "title_zh": clip.title_zh,
+        "summary_zh": clip.summary_zh,
+        "reason": clip.reason,
+        "tags": clip.tags,
+        "transcript": nearby_transcript,
+    }
+    logger.info("thumbnail command start clip_id=%s command=%s output=%s", clip.id, settings.thumbnail_command, output_path)
+    completed = subprocess.run(
+        shlex.split(settings.thumbnail_command),
+        input=json.dumps(payload, ensure_ascii=False),
+        text=True,
+        capture_output=True,
+        check=False,
+    )
+    if completed.returncode != 0:
+        logger.error("thumbnail command failed clip_id=%s stderr=%s", clip.id, completed.stderr[-4000:])
+        raise RuntimeError(completed.stderr)
+    if not output_path.exists():
+        raise RuntimeError(f"Thumbnail command did not create {output_path}")
+    logger.info("thumbnail command done clip_id=%s stdout=%s", clip.id, completed.stdout[-1000:])
+    return output_path
+
+
+def compose_thumbnail(
+    *,
+    background_path: Path,
+    output_path: Path,
+    title: str,
+    character_overlay_path: Path | None,
+    character_scale: float,
+    character_position: str,
+    title_enabled: bool,
+    width: int,
+    height: int,
+) -> None:
+    from PIL import Image, ImageDraw, ImageEnhance, ImageFont, ImageOps
+
+    background = Image.open(background_path).convert("RGB")
+    canvas = ImageOps.fit(background, (width, height), method=Image.Resampling.LANCZOS)
+    canvas = ImageEnhance.Color(canvas).enhance(1.12)
+    canvas = ImageEnhance.Contrast(canvas).enhance(1.08)
+    rgba = canvas.convert("RGBA")
+
+    if title_enabled and title.strip():
+        draw_title(rgba, title)
+
+    if character_overlay_path:
+        overlay_character(rgba, character_overlay_path, character_scale, character_position)
+
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    rgba.convert("RGB").save(output_path, quality=92, optimize=True)
+
+
+def draw_title(image, title: str) -> None:
+    from PIL import Image, ImageDraw
+
+    draw = ImageDraw.Draw(image)
+    width, height = image.size
+    font = load_font(max(48, width // 18))
+    max_text_width = int(width * 0.62)
+    lines = wrap_text(draw, title, font, max_text_width, max_lines=3)
+    line_boxes = [draw.textbbox((0, 0), line, font=font, stroke_width=2) for line in lines]
+    line_height = max((box[3] - box[1] for box in line_boxes), default=font.size)
+    block_height = line_height * len(lines) + 18 * max(0, len(lines) - 1)
+    left = int(width * 0.045)
+    top = int(height * 0.075)
+    padding_x = 34
+    padding_y = 26
+    block_width = min(max_text_width + padding_x * 2, int(width * 0.72))
+
+    panel = Image.new("RGBA", (block_width, block_height + padding_y * 2), (0, 0, 0, 150))
+    image.alpha_composite(panel, (left - padding_x, top - padding_y))
+
+    y = top
+    for line in lines:
+        draw.text(
+            (left, y),
+            line,
+            font=font,
+            fill=(255, 255, 255, 255),
+            stroke_width=3,
+            stroke_fill=(0, 0, 0, 210),
+        )
+        y += line_height + 18
+
+
+def overlay_character(image, overlay_path: Path, scale: float, position: str) -> None:
+    from PIL import Image
+
+    if not overlay_path.exists():
+        logger.warning("thumbnail character overlay missing path=%s", overlay_path)
+        return
+    overlay = Image.open(overlay_path).convert("RGBA")
+    width, height = image.size
+    target_height = int(height * scale)
+    ratio = target_height / overlay.height
+    target_size = (max(1, int(overlay.width * ratio)), target_height)
+    overlay = overlay.resize(target_size, Image.Resampling.LANCZOS)
+
+    margin_x = int(width * 0.035)
+    margin_y = int(height * 0.02)
+    positions = {
+        "bottom-right": (width - overlay.width - margin_x, height - overlay.height - margin_y),
+        "bottom-left": (margin_x, height - overlay.height - margin_y),
+        "center-right": (width - overlay.width - margin_x, (height - overlay.height) // 2),
+        "center-left": (margin_x, (height - overlay.height) // 2),
+    }
+    x, y = positions.get(position, positions["bottom-right"])
+    image.alpha_composite(overlay, (max(0, x), max(0, y)))
+
+
+def load_font(size: int):
+    from PIL import ImageFont
+
+    for candidate in FONT_CANDIDATES:
+        path = Path(candidate)
+        if path.exists():
+            try:
+                return ImageFont.truetype(str(path), size=size)
+            except OSError:
+                continue
+    return ImageFont.load_default(size=size)
+
+
+def wrap_text(draw, text: str, font, max_width: int, max_lines: int) -> list[str]:
+    lines: list[str] = []
+    current = ""
+    for char in text.strip():
+        candidate = current + char
+        bbox = draw.textbbox((0, 0), candidate, font=font, stroke_width=2)
+        if bbox[2] - bbox[0] <= max_width or not current:
+            current = candidate
+            continue
+        lines.append(current)
+        current = char
+        if len(lines) == max_lines:
+            break
+    if current and len(lines) < max_lines:
+        lines.append(current)
+    if len(lines) == max_lines and len("".join(lines)) < len(text.strip()):
+        lines[-1] = lines[-1].rstrip("，。,. ") + "..."
+    return lines or [text.strip()]
@@ -0,0 +1,86 @@
+from __future__ import annotations
+
+import json
+import logging
+import subprocess
+from pathlib import Path
+
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import Artifact, ClipSuggestion
+
+logger = logging.getLogger(__name__)
+
+
+class UploadResult(dict):
+    pass
+
+
+def upload_artifact(session: Session, clip: ClipSuggestion, artifact: Artifact) -> UploadResult:
+    settings = get_settings()
+    thumbnail = (
+        session.query(Artifact)
+        .filter(Artifact.clip_id == clip.id, Artifact.artifact_type == "thumbnail_final")
+        .order_by(Artifact.created_at.desc())
+        .first()
+    )
+    if settings.upload_adapter == "noop":
+        logger.info(
+            "upload noop clip_id=%s artifact_id=%s path=%s thumbnail=%s",
+            clip.id,
+            artifact.id,
+            artifact.local_path,
+            thumbnail.local_path if thumbnail else None,
+        )
+        return UploadResult(
+            {
+                "adapter": "noop",
+                "uploaded": False,
+                "path": artifact.local_path,
+                "thumbnail_path": thumbnail.local_path if thumbnail else None,
+            }
+        )
+    if settings.upload_adapter != "command":
+        raise RuntimeError(f"Unknown upload adapter: {settings.upload_adapter}")
+    if not settings.upload_command:
+        raise RuntimeError('[upload].command must be set when [upload].adapter = "command"')
+
+    payload = {
+        "clip_id": clip.id,
+        "video_id": clip.video_id,
+        "title": clip.title_zh,
+        "summary": clip.summary_zh,
+        "tags": clip.tags,
+        "file": artifact.local_path,
+        "thumbnail": thumbnail.local_path if thumbnail else None,
+    }
+    logger.info("upload command start clip_id=%s artifact_id=%s command=%s", clip.id, artifact.id, settings.upload_command)
+    completed = subprocess.run(
+        settings.upload_command.split(),
+        input=json.dumps(payload, ensure_ascii=False),
+        text=True,
+        capture_output=True,
+        check=False,
+    )
+    if completed.returncode != 0:
+        logger.error("upload command failed clip_id=%s stderr=%s", clip.id, completed.stderr)
+        raise RuntimeError(completed.stderr)
+    logger.info("upload command done clip_id=%s stdout=%s", clip.id, completed.stdout[-1000:])
+    return UploadResult(
+        {
+            "adapter": "command",
+            "uploaded": True,
+            "path": artifact.local_path,
+            "stdout": completed.stdout,
+        }
+    )
+
+
+def choose_upload_artifact(artifacts: list[Artifact]) -> Artifact | None:
+    by_type = {artifact.artifact_type: artifact for artifact in artifacts}
+    return by_type.get("clip_baked_sub_mp4") or by_type.get("clip_soft_sub_mp4")
+
+
+def artifact_path(artifact: Artifact) -> Path:
+    return Path(artifact.local_path)
@@ -0,0 +1,177 @@
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+from datetime import UTC, datetime
+from pathlib import PurePosixPath
+from typing import Any
+from urllib.parse import quote, urljoin, urlparse
+from xml.etree import ElementTree
+
+import httpx
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import Video
+
+VIDEO_SUFFIXES = {".flv", ".mp4", ".mkv", ".mov"}
+logger = logging.getLogger(__name__)
+
+
+@dataclass(frozen=True)
+class WebDavFile:
+    url: str
+    filename: str
+    size_bytes: int
+    modified_at: str | None = None
+
+
+class WebDavClient:
+    def __init__(
+        self,
+        base_url: str,
+        username: str | None = None,
+        password: str | None = None,
+        verify_tls: bool = True,
+    ) -> None:
+        self.base_url = str(base_url).rstrip("/") + "/"
+        self.auth = (username, password) if username and password else None
+        self.verify_tls = verify_tls
+
+    @classmethod
+    def from_settings(cls) -> WebDavClient:
+        settings = get_settings()
+        return cls(
+            str(settings.webdav_base_url),
+            settings.webdav_username,
+            settings.webdav_password,
+            settings.webdav_verify_tls,
+        )
+
+    def list_files(self) -> list[WebDavFile]:
+        logger.debug("webdav propfind start base_url=%s", self.base_url)
+        body = """<?xml version="1.0" encoding="utf-8" ?>
+<D:propfind xmlns:D="DAV:">
+  <D:prop>
+    <D:getcontentlength/>
+    <D:getlastmodified/>
+    <D:resourcetype/>
+  </D:prop>
+</D:propfind>"""
+        with httpx.Client(auth=self.auth, verify=self.verify_tls, timeout=30) as client:
+            response = client.request("PROPFIND", self.base_url, headers={"Depth": "1"}, content=body)
+            response.raise_for_status()
+        files = parse_propfind_response(response.text, self.base_url)
+        logger.info("webdav propfind done base_url=%s files=%s", self.base_url, len(files))
+        return files
+
+    def download(self, url: str, destination) -> None:
+        logger.info("webdav download start url=%s destination=%s", url, destination)
+        with httpx.stream("GET", url, auth=self.auth, verify=self.verify_tls, timeout=None) as response:
+            response.raise_for_status()
+            with open(destination, "wb") as out_file:
+                for chunk in response.iter_bytes():
+                    out_file.write(chunk)
+        logger.info("webdav download done url=%s destination=%s", url, destination)
+
+    def upload(self, source, relative_path: str) -> str:
+        destination = urljoin(self.base_url, quote(relative_path.lstrip("/")))
+        logger.info("webdav upload start source=%s destination=%s", source, destination)
+        with open(source, "rb") as in_file:
+            response = httpx.put(
+                destination,
+                content=in_file,
+                auth=self.auth,
+                verify=self.verify_tls,
+                timeout=None,
+            )
+        response.raise_for_status()
+        logger.info("webdav upload done source=%s destination=%s", source, destination)
+        return destination
+
+
+def parse_propfind_response(xml_text: str, base_url: str) -> list[WebDavFile]:
+    ns = {"d": "DAV:"}
+    root = ElementTree.fromstring(xml_text)
+    base_path = urlparse(base_url).path.rstrip("/")
+    files: list[WebDavFile] = []
+    for response in root.findall("d:response", ns):
+        href = response.findtext("d:href", default="", namespaces=ns)
+        href_path = urlparse(href).path
+        if href_path.rstrip("/") == base_path:
+            continue
+        filename = PurePosixPath(href_path).name
+        if not filename or PurePosixPath(filename).suffix.lower() not in VIDEO_SUFFIXES:
+            continue
+        resource_type = response.find(".//d:resourcetype", ns)
+        if resource_type is not None and list(resource_type):
+            continue
+        size_text = response.findtext(".//d:getcontentlength", default="0", namespaces=ns)
+        modified = response.findtext(".//d:getlastmodified", default=None, namespaces=ns)
+        file_url = urljoin(base_url.rstrip("/") + "/", quote(filename))
+        files.append(WebDavFile(file_url, filename, int(size_text or 0), modified))
+    return files
+
+
+def add_size_sample(video: Video, size_bytes: int) -> None:
+    samples: list[dict[str, Any]] = list(video.size_samples or [])
+    samples.append({"at": datetime.now(UTC).isoformat(), "size_bytes": size_bytes})
+    video.size_samples = samples[-10:]
+    video.size_bytes = size_bytes
+
+
+def has_stable_size(video: Video, required_equal_samples: int = 2) -> bool:
+    samples = video.size_samples or []
+    if len(samples) < required_equal_samples:
+        return False
+    recent = samples[-required_equal_samples:]
+    sizes = {sample["size_bytes"] for sample in recent}
+    return len(sizes) == 1 and next(iter(sizes)) > 0
+
+
+def bootstrap_existing(client: WebDavClient, session: Session) -> int:
+    count = 0
+    for file in client.list_files():
+        video = session.query(Video).filter(Video.source_url == file.url).one_or_none()
+        if video is None:
+            video = Video(
+                source_url=file.url,
+                filename=file.filename,
+                size_bytes=file.size_bytes,
+                size_samples=[{"at": datetime.now(UTC).isoformat(), "size_bytes": file.size_bytes}],
+                ingest_status="existing_done",
+                processing_status="done",
+            )
+            session.add(video)
+            count += 1
+            logger.debug("bootstrap existing file=%s size=%s", file.filename, file.size_bytes)
+    session.flush()
+    logger.info("bootstrap existing done inserted=%s", count)
+    return count
+
+
+def scan_once(client: WebDavClient, session: Session) -> tuple[int, int]:
+    observed = 0
+    newly_stable = 0
+    for file in client.list_files():
+        observed += 1
+        video = session.query(Video).filter(Video.source_url == file.url).one_or_none()
+        if video is None:
+            logger.info("new webdav file observed file=%s size=%s", file.filename, file.size_bytes)
+            video = Video(
+                source_url=file.url,
+                filename=file.filename,
+                ingest_status="observing",
+                processing_status="pending",
+            )
+            session.add(video)
+        if video.ingest_status in {"existing_done", "stable", "queued"}:
+            continue
+        add_size_sample(video, file.size_bytes)
+        logger.debug("webdav size sample video_id=%s file=%s size=%s", video.id, file.filename, file.size_bytes)
+        if has_stable_size(video):
+            video.ingest_status = "stable"
+            newly_stable += 1
+            logger.info("webdav file stable video_id=%s file=%s size=%s", video.id, file.filename, file.size_bytes)
+    session.flush()
+    return observed, newly_stable
@@ -0,0 +1,39 @@
+from sqlalchemy.orm import Session
+
+from evanescere.config import get_settings
+from evanescere.models import Setting
+from evanescere.schemas import SettingsPatch, SettingsRead
+
+SETTINGS_KEY = "pipeline"
+
+
+def default_settings() -> SettingsRead:
+    settings = get_settings()
+    return SettingsRead(
+        suggest_enabled=settings.default_suggest_enabled,
+        render_enabled=settings.default_render_enabled,
+        upload_enabled=settings.default_upload_enabled,
+        preserve_final_artifacts=settings.default_preserve_final_artifacts,
+        bake_subtitles=settings.default_bake_subtitles,
+    )
+
+
+def get_pipeline_settings(session: Session) -> SettingsRead:
+    row = session.get(Setting, SETTINGS_KEY)
+    if row is None:
+        return default_settings()
+    return SettingsRead(**(default_settings().model_dump() | row.value))
+
+
+def patch_pipeline_settings(session: Session, patch: SettingsPatch) -> SettingsRead:
+    current = get_pipeline_settings(session).model_dump()
+    current.update({k: v for k, v in patch.model_dump().items() if v is not None})
+    row = session.get(Setting, SETTINGS_KEY)
+    if row is None:
+        row = Setting(key=SETTINGS_KEY, value=current)
+        session.add(row)
+    else:
+        row.value = current
+    session.flush()
+    return SettingsRead(**current)
+
@@ -0,0 +1,64 @@
+from evanescere.schemas import ClipCandidate
+from evanescere.services.llm import dedupe_and_rank_candidates, parse_clip_response
+
+
+def test_parse_clip_response():
+    candidates = parse_clip_response(
+        """
+        {
+          "clips": [
+            {
+              "start_sec": 10,
+              "end_sec": 60,
+              "title_zh": "标题",
+              "summary_zh": "摘要",
+              "reason": "有趣",
+              "score": 0.9,
+              "tags": ["反应"],
+              "subtitle_priority": "high"
+            }
+          ]
+        }
+        """
+    )
+    assert candidates[0].title_zh == "标题"
+
+
+def test_dedupe_and_rank_candidates_filters_short_and_overlapping():
+    candidates = [
+        ClipCandidate(
+            start_sec=0,
+            end_sec=20,
+            title_zh="too short",
+            summary_zh="",
+            reason="",
+            score=1,
+        ),
+        ClipCandidate(
+            start_sec=0,
+            end_sec=60,
+            title_zh="best",
+            summary_zh="",
+            reason="",
+            score=0.9,
+        ),
+        ClipCandidate(
+            start_sec=10,
+            end_sec=65,
+            title_zh="overlap",
+            summary_zh="",
+            reason="",
+            score=0.8,
+        ),
+        ClipCandidate(
+            start_sec=120,
+            end_sec=180,
+            title_zh="second",
+            summary_zh="",
+            reason="",
+            score=0.7,
+        ),
+    ]
+    ranked = dedupe_and_rank_candidates(candidates)
+    assert [candidate.title_zh for candidate in ranked] == ["best", "second"]
+
@@ -0,0 +1,10 @@
+from evanescere.services.subtitles import format_ass_time, format_srt_time
+
+
+def test_format_srt_time():
+    assert format_srt_time(3661.234) == "01:01:01,234"
+
+
+def test_format_ass_time():
+    assert format_ass_time(3661.23) == "1:01:01.23"
+
@@ -0,0 +1,35 @@
+from evanescere.services.webdav import WebDavFile, has_stable_size, parse_propfind_response
+
+
+class DummyVideo:
+    def __init__(self, samples):
+        self.size_samples = samples
+
+
+def test_has_stable_size_requires_two_equal_positive_samples():
+    assert not has_stable_size(DummyVideo([]))
+    assert not has_stable_size(DummyVideo([{"size_bytes": 100}]))
+    assert not has_stable_size(DummyVideo([{"size_bytes": 100}, {"size_bytes": 101}]))
+    assert has_stable_size(DummyVideo([{"size_bytes": 100}, {"size_bytes": 100}]))
+
+
+def test_parse_propfind_response_filters_video_files():
+    xml = """<?xml version="1.0"?>
+<D:multistatus xmlns:D="DAV:">
+  <D:response>
+    <D:href>/webdav/recordings/</D:href>
+    <D:propstat><D:prop><D:resourcetype><D:collection/></D:resourcetype></D:prop></D:propstat>
+  </D:response>
+  <D:response>
+    <D:href>/webdav/recordings/stream.flv</D:href>
+    <D:propstat><D:prop><D:getcontentlength>123</D:getcontentlength></D:prop></D:propstat>
+  </D:response>
+  <D:response>
+    <D:href>/webdav/recordings/readme.txt</D:href>
+    <D:propstat><D:prop><D:getcontentlength>12</D:getcontentlength></D:prop></D:propstat>
+  </D:response>
+</D:multistatus>"""
+    assert parse_propfind_response(xml, "https://host/webdav/recordings") == [
+        WebDavFile("https://host/webdav/recordings/stream.flv", "stream.flv", 123, None)
+    ]
+
				`@@ -0,0 +1,2 @@`
				`"""Service adapters for storage, AI, media, subtitles, and upload."""`