What it does Jellyfish is a full-stack workspace for producing AI-generated short dramas. You feed it a script; it breaks chapters into shots, extracts characters, props, costumes, and dialogue, then walks you through preparation, generation, and export. The stack is React/Vite frontend, FastAPI backend, MySQL, Redis, and RustFS, all wired up with Docker Compose.

The interesting bit The project treats “consistency” as infrastructure, not an afterthought. Characters, scenes, props, and costumes live in a shared entity model that persists across shots—so your lead actor doesn’t mysteriously switch faces between scenes. Long-running generation jobs (text, image, video) become trackable async tasks with cancellation and recovery, which is the unglamorous plumbing that keeps batch production from becoming a debugging nightmare.

Key highlights

• Script-to-shot breakdown with AI extraction of characters, dialogue, and production elements
• Centralized asset library to reduce style drift across generated shots
• Unified async task center for tracking, canceling, and recovering generation jobs
• Multi-provider model management with prompt templates per category
• OpenAPI-driven frontend contracts with auto-generated TypeScript helpers
• Docker Compose setup with explicit ports and environment wiring

Caveats

• The README describes capabilities extensively but offers no benchmarks, sample outputs, or performance numbers
• “RustFS” appears in the compose ports but isn’t explained; presumably an S3-compatible object store, though this isn’t confirmed
• Generation quality depends entirely on whichever external image/video models you configure

Verdict Worth a look if you’re running an AI video studio or producing short dramas at volume and need guardrails against chaos. Skip it if you just want a one-click video generator—this is a workflow tool, not a magic button.