Features
End-to-end short-video generation from text prompt
Multi-model AI orchestration (DeepSeek, Qwen, GLM, Kling v2.5, faster-whisper, multiple TTS)
Circuit breaker pattern for resilient API calls
Usage metering and quota tracking per tenant
Full-stack observability (Langfuse, cross-language trace_id propagation)
Prompt anchoring for consistent visual identity
CLIP-based text-image consistency scoring and gating
Automatic audio-video synchronization rescue (4-tier strategy)
SSE progress streaming for real-time job status
Single-segment regeneration without full rebuild
Spring Boot API gateway with JWT auth and routing
FastAPI orchestration with async workers (ARQ)
Vue.js frontend (Vue Flow, Pinia, Tailwind)
Docker Compose one-click deployment
Database persistence via SQLAlchemy + Alembic migrations (PostgreSQL/Redis/MinIO)
Mochi 1 open-source text-to-video model
High-fidelity motion with strong prompt adherence
Interactive playground for prompt testing
Open-source code on GitHub and Hugging Face
ComfyUI integration for custom workflows
Command-line interface for batch video generation
Quickstart script (quickstart.sh) for local setup
Credit-based usage (Free, Lite, Standard tiers)
Mochi video generation costs 100 credits
Replay video generation costs 10 credits
No watermark on paid tiers
Commercial usage rights on paid tiers
Early model access on Standard tier
Active research blog with model updates
