Anil-matcha/Open-Generative-AI

What it does

Open Generative AI is a Next.js-based creative studio that puts many third-party image, video, audio and lip-sync models behind one interface. It is aimed at people who want a self-hostable front end for model providers, rather than a single model implementation: generation requests are routed through Muapi.ai, and the UI exposes separate Image, Video, Audio, Lip Sync, Cinema, Workflow and Agent studios.

The repository describes five generations of formats: text-to-image, image-to-image, text-to-video, image-to-video and video-to-video, plus audio and lip-sync workflows. It also documents local inference through two optional engines in the desktop app, sd.cpp for image models and a separately installed Wan2GP for video and larger image models. Local engine support is a desktop-only feature; the hosted version always uses cloud APIs.

Quick start

For the browser version, the README points to a hosted studio that requires no Node.js setup (hosted version). Sign up for a free account and enter the Muapi access key when prompted.

For a source build, you need Node.js 18+ and a Muapi.ai access key. Clone the repository with submodules, install and build the workspace packages, then choose the desktop or web entry point:

git clone --recurse-submodules https://github.com/Anil-matcha/Open-Generative-AI.git
cd Open-Generative-AI
npm run setup
npm run electron:dev   # desktop app
# or
npm run dev            # web version at http://localhost:3000

npm install alone is not sufficient; npm run setup builds the shared studio, workflow and agent packages. On first use, enter the access key value, not its label. The key is stored in browser localStorage and used for API requests to Muapi.

Prebuilt installers for macOS, Windows and Linux are listed on the release page. macOS builds are not notarized, and Windows installers are not code-signed, so the README describes Gatekeeper and SmartScreen bypass steps; treat those binaries as unaudited third-party software and prefer building from source if you need to inspect what you run.

Requirements and limits

• Node.js 18+ for source builds; the Electron app uses a monorepo with npm workspaces.
• Muapi.ai key for cloud models. Local models can be used without a key in the desktop app.
• Local inference hardware: sd.cpp runs on CPU and Apple Silicon Metal, with CUDA/Vulkan/ROCm on Linux and Windows. Z-Image models need roughly 16 GB RAM; the README warns that they can hang an 8 GB M-series Mac. Wan2GP requires a separately installed Python/PyTorch environment with a CUDA or ROCm GPU.
• Ubuntu 24.04+: the AppImage may fail because of AppArmor user-namespace restrictions. The .deb package is the recommended option; the AppImage workaround changes a system-wide sysctl setting.
• Data handling: API keys and upload history are kept in browser localStorage, while generation requests and uploaded files go to Muapi. Local model files and engine data stay in the app data directory unless OPEN_GENERATIVE_AI_LOCAL_AI_DIR is set.
• License: MIT for the repository code; model access and usage remain subject to the upstream providers.

HotRepo's take

The repository has a JavaScript-language daily snapshot from 5 November 2026 at position 3, with 32,340 stars and 123 stars gained that day. Earlier daily snapshots in the data show growth from 14,206 stars on 16 May 2026 to 21,560 on 28 June 2026, but these are separate daily records rather than a complete trend line. The project's appeal is a broad, self-hosted front end over many commercial model endpoints; whether that is useful depends heavily on the Muapi account and the bundled third-party model access, not on a model shipped by this repository.