Digital Writer · Digital Director · AI Detection · Multi-Channel Payments
中文 · 繁體中文 · 日本語 · ไทย · Tiếng Việt · Русский · Capabilities · Screenshots · Quick Start · Deployment
Huohuo Drama is an open-source, full-stack AI short-drama and AI novel production platform. One workspace takes raw text through formatted scripts, storyboards, and multimodal assets to finished episodes—with TTS, per-shot FFmpeg mux, and export.
Each episode supports two production pipelines—AI video and frame slideshow (separate workbenches, can run in parallel): the AI path uses image-to-video; the frame path builds shots from keyframe sequences with FFmpeg Ken Burns motion, aligns clip length to TTS duration, adds background music on dialogue-free shots, then muxes and merges into a full episode.
AI novel writing: the built-in Digital Writer runs brief → draft → consistency review in batch. Causal-chain mode is on by default: each chapter ends with a Change Record block where every state shift must spell out trigger → process → outcome. The system stores, injects, and audits these chains across later chapters—together with four-layer continuity memory—to keep character, realm, foreshadowing, and cross-chapter logic consistent at novel length.
Novel → short drama: import chapters or a full AI novel into a drama project, rewrite into a shootable screenplay, then run storyboard breakdown and the full video pipeline—no retyping between formats.
Built for creators and small teams, the platform productizes writing, production, quality checks, and billing:
| Capability | Description |
|---|---|
| Digital Writer | Server-side batch chapters: brief → draft → consistency review; causal-chain change records at chapter end keep long serials coherent |
| Digital Director | Server-side batch episodes: choose AI video or frame slideshow at start; script → storyboard → assets → mux → merge runs in the background with progress restore after refresh |
| AI Detection | Built-in AI text detection and de-AI rewriting (console route /ai-detect) for compliance-friendly publishing and polish |
| Multi-channel payments | Credit-based billing with Stripe, PayPal, PingPong, WeChat Pay, and Alipay—admins enable channels via env vars and the settings UI |
| Mobile command station | mobile-app/ uni-app client (WeChat mini program / Android / iOS / H5): pick projects, dispatch batch commands, track job progress; WeChat login & pay; deep-link to Web for detailed editing |
Short-drama flow: cast portraits → scene extraction → storyboard → AI image-to-video or frame Ken Burns slideshow → TTS / per-shot mux → FFmpeg episode merge & export.
Novel flow: Digital Writer batch → causal chain (chapter-end change records + causal audit) → four-layer memory injection → retrieval-augmented continuity.
Novel → short drama: import from a novel project → screenplay rewrite → storyboard & generation → episode export (same workspace, shared projects).
Mobile command station (mobile-app/): A lightweight on-the-go companion to the Web workbench—select a project, send Digital Writer / Director batch commands, and monitor progress from your phone. Chapter editing, storyboards, and mux stay on Web (workbench). One uni-app codebase targets WeChat mini program, Android, iOS, and H5 (dev :48555).
| Login | Settings |
|---|---|
 |
 |
| AI detection | Novel — chapter list |
|---|---|
 |
 |
| Novel — chapter editor | Short drama — production workbench |
|---|---|
 |
 |
| Mobile — projects | Mobile — tasks |
|---|---|
 |
 |
| Area | What you get |
|---|---|
| Dual pipelines | Per episode: AI video workbench or frame slideshow workbench; frames use keyframe sequences + Ken Burns—no video-model credits |
| Accounts | Multi-user auth, JWT, credit ledger and usage history |
| Template gallery | Publish drama projects as reusable templates |
| Lesson library | Per-agent hints injected into prompts over time |
| Skill extensions | agent-skills/ SKILL.md playbooks; upload ZIPs in Settings |
workbench/ Nuxt 3 workbench (dev :28555)
workbench-server/ Hono API + Drizzle + Mastra agents (dev :18555)
mobile-app/ uni-app mobile command station (H5 dev :48555)
deploy/ Docker Compose + nginx (console + API)
workbench-data/ SQLite DB (default) + static media under workbench-data/static/
+ docs screenshots under workbench-data/images/
agent-skills/ Agent SKILL.md playbooks (uploadable in Settings)
desktop/ Optional Electron shell (hosted console only)
- Cast & locations — AI portraits, uploads, voice assignment and preview
- Storyboard — Shot breakdown, prompts, stills / keyframe sequences, grid split/assign
- Video & audio — AI image-to-video, or frame Ken Burns slideshow; TTS (frame clips follow dub duration), per-shot FFmpeg mux, async episode merge
- Media library — Local storage with async job progress
| Agent id | Role |
|---|---|
drama_script_formatter |
Raw prose → formatted shooting script |
drama_cast_scene_extract |
Cast + location extraction |
drama_storyboard_breakdown |
Script → ordered shot list |
drama_voice_assign |
Map voices to cast members |
drama_image_prompt |
Prompt packs for cast, scenes, grid frames |
| Modality | Vendors |
|---|---|
| Text | OpenAI, Gemini, DeepSeek, GLM, MiniMax, Volcengine, Ali, OpenRouter |
| Image | OpenAI, Gemini, MiniMax, Volcengine, Ali |
| Video | MiniMax, Volcengine/Seedance, Vidu, Ali |
| TTS | MiniMax |
Long-form projects use four-layer continuity memory: a global state snapshot, previous-chapter tail, earlier summaries, and keyword-retrieved ledgers — all injected with hard size caps so prompts stay bounded as chapter count grows. Batch writing supports brief → chapter → consistency check; strict mode can loop local fixes until checks pass.
Digital writer / director jobs run on the server after the client disconnects. Requires login (JWT). One active job per drama; progress restores via GET /api/v1/batch-jobs/active.
The desktop/ package is a thin Electron wrapper for the hosted console. It does not bundle the local API. See desktop/README.md.
| Tool | Minimum | Notes |
|---|---|---|
| Node.js | 22+ | API and Nuxt dev servers |
| npm | 9+ | Package manager |
| FFmpeg | 4.0+ | Required for mux/concat |
# macOS: brew install ffmpeg
# Ubuntu: sudo apt install ffmpeg
ffmpeg -versioncp deploy/config.example.yaml deploy/config.yaml # optional; AI defaults, not DB driver
cp workbench-server/.env.example workbench-server/.env # authoritative for DB| Variable | Default | Purpose |
|---|---|---|
DB_DRIVER |
sqlite |
sqlite or mysql |
DB_PATH |
workbench-data/huohuo_drama.db |
SQLite file |
DATABASE_URL |
— | MySQL DSN |
DB_AUTO_INIT |
true |
Auto DDL + seed on boot |
PORT |
18555 |
API listen port |
Provider keys and models are configured in the web Settings UI.
git clone https://github.com/appolloqin/huohuo-drama.git
cd huohuo-drama
cd workbench-server && npm install
cd ../workbench && npm install
# terminal A
cd workbench-server && npm run dev
# terminal B
cd workbench && npm run dev- Console:
http://localhost:28555(preferlocalhostover127.0.0.1on some Windows setups) - API:
http://localhost:18555/api/v1(Nuxt dev proxy for/apiand/static)
Single process (production-like locally):
cd workbench && npm run generate
cd ../workbench-server && npm start
# → http://localhost:18555Smoke checks:
cd workbench-server && npm run smoke:flow
cd ../workbench && npm run smoke:proxySQLite (default) — workbench-data/huohuo_drama.db created on first boot when DB_AUTO_INIT=true.
MySQL — in workbench-server/.env:
DB_DRIVER=mysql
DATABASE_URL=mysql://user:pass@127.0.0.1:3306/huohuo_dramaSeed MiniMax voices after first boot: cd workbench-server && npm run seed:voices
Console + API (default SQLite, port 80):
cd deploy
cp .env.example .env # optional
docker compose up -d --buildSet DB_DRIVER, DATABASE_URL, payment keys, etc. in deploy/.env — see deploy/.env.example.
Docker + MySQL (remote or self-hosted instance):
# deploy/.env
DB_DRIVER=mysql
DATABASE_URL=mysql://user:pass@host:3306/huohuo_dramacd deploy && docker compose up -d --buildDB_AUTO_INIT |
Behavior |
|---|---|
true (default) |
Create tables, patch columns, seed reference data |
false |
Connect only — no automatic DDL |
Workbench-server env: STRIPE_*, PAYPAL_*, PINGPONG_*, etc., plus SITE_URL (public origin for redirects). Webhook example: https://your-domain.com/api/v1/payments/stripe/webhook. Enable channels in Settings → Payments (Stripe / PayPal / PingPong / WeChat / Alipay).
cd workbench && npm run generate # → workbench/.output/public
cd ../workbench-server && npm startMount workbench-data/ for DB and workbench-data/static/ for media. Nginx sample: deploy/nginx.conf (console at /console/).
| Layer | Choices |
|---|---|
| API | Node.js 22+, Hono |
| DB | Drizzle ORM, SQLite (default) or MySQL 8+ via repositories |
| Agents | Mastra + AI SDK (OpenAI-compatible) |
| Media | FFmpeg, Sharp |
| UI | Nuxt 3 SPA, Vue 3, TypeScript |
Docker → host Ollama: Base URL http://host.docker.internal:11434/v1; host must listen on 0.0.0.0. On Linux docker run, add --add-host=host.docker.internal:host-gateway.
FFmpeg missing: Install and verify ffmpeg -version. Docker images include FFmpeg.
Workbench cannot reach API: Ensure workbench-server is on :18555; dev proxy is in workbench/nuxt.config.ts.
Tables not created: Set DB_AUTO_INIT=true; check logs for SQLite/MySQL connection messages.
Production MySQL: DB_DRIVER=mysql + DATABASE_URL; keep workbench-data/static on a volume.
API keys: Model aggregation portal
cd workbench-server && npm run typecheck && npm run check:layers
cd ../workbench && npm run buildCI: .github/workflows/workbench-server-ci.yml (typecheck, layer checks, SQLite/MySQL smoke).