The unfiltered AI media studio you can actually own

What it does

Open Generative AI is an Electron-based desktop app (plus hosted web version) that bundles image generation, video generation, lip sync, and a “cinema” workflow into one interface. It talks to ~200 cloud models via MuAPI — Flux, Kling, Wan 2.2, various Midjourney-alikes — and also supports two local inference paths: a bundled sd.cpp engine for images on Apple Silicon, and a bring-your-own Wan2GP server for video on CUDA/ROCm hardware.

The interesting bit

The architecture is deliberately split: the desktop app can run on a Mac while offloading heavy video inference to a Linux GPU box or rented instance. The README is unusually honest about what breaks — unsigned installers, AppArmor sandboxing on Ubuntu 24.04, 8 GB Macs hanging on Z-Image — which suggests the project has actually been battle-tested outside the author’s machine.

Key highlights

• No content filters — explicit marketing point, for better or worse
• Local + remote hybrid — sd.cpp bundled for images, Wan2GP server for video; same UI toggles between them
• 200+ cloud models via MuAPI, including text-to-video, image-to-video, and audio-driven lip sync
• Multi-image input — up to 14 reference images for compatible models
• Self-hostable — MIT licensed, though the hosted version requires a free account
• Extensible model list — add custom models and modify the UI

Caveats

• Local video inference is desktop-only and requires a separate Python/PyTorch server; the web version always hits cloud APIs
• macOS and Windows installers are unsigned, so Gatekeeper/SmartScreen warnings are expected
• Z-Image models need 16 GB RAM and are known to hang 8 GB Apple Silicon machines

Verdict

Worth a look if you want a unified generative media UI without platform lock-in, and you’re comfortable trading the safety rails of commercial tools for creative control. Skip it if you need turnkey local video on a Mac, or if “no guardrails” sounds like a liability rather than a feature.