An end-to-end automated short-video production pipeline. One command in, a publish-ready video out.

English | 简体中文

myAiVideos is an open-source engineering project for automated Chinese-language short-video production, covering the full pipeline from topic discovery → creative generation → visuals → audio → post-production → distribution. It uses FastAPI as the orchestration core and a Java gateway as the platform governance layer (auth / routing / circuit breaker / metering / observability), coordinating multiple AI models for generation and quality gating.

Suitable content formats:

• Hot-news commentary
• Knowledge explainers
• Emotional storytelling
• Curiosity / strange facts
• Social insight clips

• Seven-layer pipeline architecture: Topic (L1) → Creative (L2) → Visual (L3) → Audio (L4) → Post-production (L5) → Distribution (L6) → Optimization (L7). Layers are decoupled and independently replaceable.
• Multi-model failover + circuit breaker: Java gateway aggregates DeepSeek / Qwen / GLM with Resilience4j protection; failed providers are automatically rotated.
• AI consistency governance
  • Prompt anchoring: maintain visual identity of subjects across video segments
  • CLIP text-image consistency gating: reject off-prompt keyframes early
  • Audio-video sync auto-rescue: 4-tier strategy (audio tempo / video pad / narration rewrite)
• Full-stack observability: trace_id propagated across Java/Python; Langfuse call-tree visibility.
• Metering & rate limiting: AOP aspect captures token / cost without code changes, aggregated by tenant.
• Production-ready: SSE progress streaming, single-segment regeneration, rolling log archives.

                +----------------+
                |  Client / Bot  |
                +--------+-------+
                         |
                +--------v-------+
                |  Java Gateway  |  JWT / TraceId / Router / Breaker / Meter
                +--------+-------+
                         |
                +--------v-------+
                | FastAPI orch.  |  webhook · session · job orchestration
                +--+---------+---+
                   |         |
        +----------v---+   +-v-----------+
        |  ARQ Worker  |   |  7-layer    |
        +-----+--------+   |  pipeline   |
              |            +-+-----------+
              |              |
   +----------v--+    +------v-----------------+
   | Redis(queue)|    | DeepSeek / Kling / TTS |
   +-------------+    +------------------------+

For details, see architecture.md.

• Docker Desktop ≥ 24
• Node.js ≥ 20 (only for frontend development)
• Python ≥ 3.11 (only if running locally without Docker, or running tests on host)

# 1. Clone and prepare env
git clone <your-fork>.git
cd myAiVideos
cp .env.example .env
# Edit .env to fill in at least the required AI provider API keys.

# 2. Start backend (postgres + redis + orchestrator + worker)
docker compose up -d postgres redis orchestrator worker

# 3. Run database migrations
docker compose exec -T orchestrator alembic upgrade head

# 4. (Optional) Start Java gateway
docker compose up -d gateway

# 5. Start frontend
cd sl-vue && npm install && npm run dev

Health checks:

• Python API: http://localhost:8000/health
• Java Gateway: http://localhost:8080/actuator/health
• Frontend: http://localhost:5173

For Windows / macOS / Linux specific scripts, scenario presets, and troubleshooting, see docs/启动-停止脚本.md.

Core environment variables (full list in .env.example):

myAiVideos/
├── api/                  FastAPI routes (webhook, storyboard, clip)
├── config/               Settings, Skill templates, characters, styles
├── core/                 trace context, langfuse client, logging, scheduler
├── db/                   SQLAlchemy models, Alembic migrations
├── integrations/         External model client wrappers
├── layers/               Seven-layer pipeline
│   ├── L1_trending/      Topic discovery / ranking
│   ├── L2_creative/      Script generation, prompt orchestration, anchoring
│   ├── L3_visual/        Text-to-image, image-to-video, CLIP consistency
│   ├── L4_audio/         TTS, mixing, duration planning
│   ├── L5_postprod/      AV sync, rescue, captions, cover
│   ├── L6_distribution/  Delivery back to client
│   └── L7_optimization/  Quality scoring, data feedback
├── scripts/              Ops & backfill scripts
├── sl-vue/               Vue 3 frontend
├── gateway/              Java Spring Boot gateway (see gateway/README.md)
├── tests/                pytest test suite
├── docs/                 Module reference docs
├── docker-compose.yaml   Multi-container orchestration
├── Dockerfile            Python image
└── main_v2.py            FastAPI entrypoint

• architecture.md — Overall architecture and data flow
• docs/operations-manual.md — Operations manual: Feishu integration, testing, daily maintenance (中文: docs/操作使用手册.md)
• docs/启动-停止脚本.md — Cross-platform start / stop / log / troubleshooting guide
• gateway/README.md — Java gateway (auth, routing, failover, circuit breaker, metering)

docker compose exec -T orchestrator pytest tests/ -q

docker compose build orchestrator
docker compose up -d --force-recreate --no-build orchestrator worker

Note: docker compose restart does not rebuild the image. Use the two commands above for source code changes.

docker compose exec -T orchestrator alembic upgrade head
# Create a new migration
docker compose exec -T orchestrator alembic revision -m "your message"

Issues and pull requests are welcome. For substantial changes, please open an issue first to discuss the design.

MIT — see LICENSE.

This project integrates several excellent open-source models and commercial AI services, including but not limited to: DeepSeek, Zhipu GLM, Kling AI, Volcengine TTS, faster-whisper, Langfuse, Resilience4j.