Open Generative AI

Open Generative AI is an all-in-one open-source studio for AI image and video generation. Available as a web app, Electron desktop client, or remote GPU server, it packs 200+ cutting-edge generative models (Florence, Midjourney, Kling, Sora, Veo, Seedance, Wan, Hunyuan, LTX, Hailuo, Grok Imagine, and others) into a single UI. There is no built-in content filter, it can be self-hosted, and it is MIT-licensed. Cloud inference is provided by MuAPI; the project also supports sd.cpp (local Metal/CUDA/Vulkan) and Wan2GP remote Gradio servers, so Mac and Windows users can run locally, and non-Mac GPU compute can be offloaded over the LAN or to a cloud box.

Core capabilities break down into seven studios. First, Image Studio: with no reference image it routes to 50+ text-to-image models; with an upload it switches to 55+ image-to-image models, with multi-image batch support (up to 14 images, e.g., Nano Banana 2 Edit, Flux Kontext, GPT-4o Edit). Second, Video Studio: defaults to 40+ text-to-video models (Kling, Sora, Veo, Wan, Seedance 2.0, Hailuo, Runway, and more); once you upload a first frame it switches to 60+ image-to-video models. Third, Lip Sync Studio: nine models covering both image-plus-audio to talking-video and existing-video-plus-audio to lip-re-syncing (Infinite Talk, LTX-2.3, LatentSync, Sync Lipsync, and others). Fourth, Cinema Studio: modular cinema cameras, anamorphic lenses, focal length, and aperture presets that auto-compose into prompt modifiers. Fifth, Workflow Studio: a node editor chains multiple models into visual pipelines, with templates and a community marketplace; the underlying Vibe-Workflow engine is also open source. Sixth, Audio Studio (voice cloning, music, narration), Clipping Studio (smart slicing), and Marketing Studio (video kits). Seventh, Agent Studio / Design Agent Studio / Apps Studio / MCP & CLI — Claude Code and other coding agents can drive the 200+ models directly via Skills.

Use cases include small teams shipping short-form video assets across platforms in one go, independent creators using Midjourney or Sora equivalents without subscriptions, marketing teams generating multi-SKU videos in batch, and agent developers plugging the model set into end-to-end pipelines. Local entry point: macOS Apple Silicon with 8GB RAM can run SD 1.5 (2.1GB); video models need an external CUDA or ROCm GPU running Wan2GP, so Mac users typically offload to remote servers or providers like RunPod and vast.ai. The cloud mode requires no API key to register and try; full functionality requires a MuAPI key of your own. Licensed under MIT. GitHub currently shows 22k+ stars and 3.7k+ forks; the repository was created in May 2023 and v2.0.0 (with Audio, Vibe Motion, Design Agent, and more) shipped in May 2026.