Hey creators,
A few weeks ago, I launched a call for beta testers for a project called « MLX-Diffusion ». The goal was simple: bring state-of-the-art, local AI image generation to Apple Silicon without the headaches of complex node-based UIs or Python errors.
The response was incredible, and the project has grown so much that it needed a proper identity. Today, I am thrilled to introduce DiffusionBear! 🕶️🐻
Why DiffusionBear? If you’re on a Mac (M1/M2/M3/M4/M5/M6/Who knows) and frustrated by cloud credits, slow generations, or lack of BUZZ or tools that aren’t optimized for your hardware, this is for you. DiffusionBear is a 100% native macOS studio built on Apple’s MLX framework.
✨ What’s inside the v0.3.8 Beta:
Massive Models, Minimal RAM: Thanks to 4-bit quantization, you can run 13B parameter models like FLUX.2, Krea 2 Turbo, Z-Image-Turbo, Qwen-Image-2.1-Turbo smoothly on a standard 16GB Mac.
Zero Cloud, Total Privacy: No accounts, no uploads, no telemetry. Your prompts and images never leave your machine.
Built-in Civitai & Hugging Face Integration: Browse, download, and manage your favorite checkpoints and LoRAs directly within the app.
Local Prompt Enhancer: A built-in Qwen language model rewrites and optimizes your prompts on the fly.
We Need You! (Beta Testers Wanted) We are pushing the limits of what Apple Silicon can do, and we need your feedback to make DiffusionBear the ultimate local studio. Whether you are a casual creator or a workflow power-user, come test the beta!
📥 Download the Beta: https://github.com/OuincheWinch/DiffusionBear/releases
🌐 Learn More: https://www.diffusionbear.com
DiffusionBear-0.3.8-arm64.zip -> Unzip compiled application, move it to your /Applications directory, download you first model ( Flux.2[Klein] 4/9B is a Pure GEM , and support up to 10 reference images )
Drop your feedback, feature requests, or your best generated images in the comments below! Let’s build the best Mac AI community together.
