Someone just open-sourced a full AI cinema and image studio with 200+ models. Lip sync. Text-to-video. Cinema controls. All self-hosted. MIT license.
The project is called Open Generative AI, and it hit 5.5K GitHub stars within days of going viral. It is, by any measure, the most ambitious open-source creative AI tool released this year — and it might be the one that kills Higgsfield AI’s business model.
🎬 What It Actually Does
Open Generative AI is a single app with five studios:
Image Studio — 50+ text-to-image models and 55+ image-to-image models. Flux, Midjourney v7, GPT-4o image generation, Seedream 5.0, Ideogram v3, Nano Banana 2. Upload a reference image and it switches to edit mode automatically. Supports up to 14 reference images for multi-image editing.
Video Studio — 40+ text-to-video and 60+ image-to-video models. Kling v3, Sora 2, Veo 3, Wan 2.6, Seedance 2.0, Runway Gen-3. Upload a start frame and it animates it. Duration controls, aspect ratios, quality settings.
Lip Sync Studio — 9 dedicated models for animating portraits or syncing lips on video. Upload a face and audio, get a talking video. Infinite Talk, Wan 2.2 Speech-to-Video, LTX 2.3 Lipsync, LatentSync. This is the feature Higgsfield AI built their brand on. Now it’s free and open.
Cinema Studio — Photorealistic cinematic shots with professional camera controls. Lens selection, focal length, aperture simulation. Modular 8K Digital, Full-Frame Cine, Grand Format 70mm, Classic 16mm. This isn’t a filter — it’s prompt engineering with cinema-specific modifiers.
Workflow Studio — Node-based visual pipeline builder. Chain image, video, and audio models into automated flows. Browse community templates or build your own.
💰 Why This Matters: The Price Gap
Higgsfield AI charges subscription fees for a subset of these capabilities behind content filters and platform restrictions. Open Generative AI gives you:
- 200+ models vs Higgsfield’s curated selection
- No content filters — no prompt rejections, no guardrails
- No subscription — free, MIT licensed
- Self-hosted — your data stays on your machine
- Multi-image input — up to 14 reference images (Higgsfield doesn’t support this)
- Lip sync included — the feature Higgsfield built its brand on, now free
- Desktop apps — macOS, Windows, Linux installers with one click
The only cost is the Muapi.ai API key for model inference. You can self-host the models too if you have the GPU hardware — which, as we covered yesterday, is now shockingly accessible.
🔍 The Technical Details
The app is a Next.js 14 monorepo with a shared React component library. You can run it with:
git clone https://github.com/Anil-matcha/Open-Generative-AI.git
cd Open-Generative-AI
npm install
npm run dev
Or download the desktop app for macOS (Apple Silicon + Intel), Windows, or Linux. One click, no terminal needed.
The architecture is clean: submit generation requests to the Muapi.ai API, poll for results. API keys stored in localStorage. Generation history persists across sessions. Every model category has intelligent mode switching — upload an image and the UI adapts.
🤔 The Catch
The project relies on Muapi.ai as its inference backend. That’s not fully self-hosted — you need an API key and the models run on Muapi’s servers. For truly local inference, you’d need GPU hardware and a compatible inference server.
But here’s the thing: the app is MIT licensed. The React components are in a shared packages/studio library. Anyone can fork it, swap Muapi for local inference (Ollama, vLLM, SGLang), and have a fully offline creative suite. The hard part — the UI, the model routing, the studio workflows — is done.
And with Qwen3-Coder 8B running locally on any 8GB GPU, the infrastructure for self-hosted creative AI is rapidly becoming trivial.
📊 The Numbers
| Metric | Value |
|---|---|
| GitHub stars | 5,800+ (and climbing) |
| License | MIT |
| Text-to-image models | 50+ |
| Image-to-image models | 55+ |
| Text-to-video models | 40+ |
| Image-to-video models | 60+ |
| Lip sync models | 9 |
| Cinema camera systems | 6 |
| Multi-image input | Up to 14 references |
| Desktop platforms | macOS, Windows, Linux |
🔍 THE BOTTOM LINE
This is what open source does. Higgsfield AI built a business on creative AI tools with content filters and subscriptions. Someone open-sourced the same thing — more models, no filters, no cost, MIT license — and 5,800 people starred it in days.
The creative AI tool landscape has a pattern: a closed platform gets traction, then the community replicates it with more features and zero cost. Stable Diffusion did it to Midjourney. Ollama did it to OpenAI’s API. Open Generative AI is doing it to Higgsfield.
If you’re paying for AI creative tools, try this first. You might not need the subscription.
Links: GitHub → Open-Generative-AI | Try it online
