WTF?! This changes image generation forever! 🤯 PrismML just released Binary and Ternary Bonsai Image 4B! That's right, 1-bit diffusion models are here. Only ~3GB in size (FLUX.2 Klein 4B is 16GB). The most shocking part? It can run 100% locally in your browser. Try it now! 👇

Massive congrats to the @PrismML team on the release — truly a game-changer! For those interested, we load the MLX weights and use custom WebGPU kernels for inference. Also, the animated Bonsai landing page was prototyped with @omma_ai 🔗 Link to demo:

Here is their HF link as I've seen a few people asking:

Forever? this doesn't change image generation at all

@victormustar Holy shit my ears are so good I know with 100% certainty the music was generated by Suno.

this tech is actually wild, but does it really look as crisp as the larger models or is it just pixelated fluff

This is actually massive for local privacy. Getting a 4B model down to 1GB without losing much quality is the real win here. Most people don't realize how hard it is to run diffusion on a phone without heating it up. I am seeing the end of cloud dependency for basic edits only.

there's a bug right there, my cursor was here, it seems the displacement only happens on a fixed section of the voxels

The change has little to do with image generation and everything to do with file size. 1-bit diffusion in a browser is neat. But can it render hands? I'd wait for benchmarks before calling it a breakthrough.

Funny how everyone thinks the bonsai animation is the breakthrough 😆

1-bit diffusion models sound revolutionary, but local performance usually comes with limitations. Can it handle complex images well?

my browser is about to start sounding like a jet engine.

@victormustar whats the point of this where the image quality is less than the bigger models

wold love to run on confyui...

Can I download the model and weights and actually run this locally?

Flux Klein 4b is most definitely not 16GB

That's actually crazy, can't imagine where we will be year.

@trentster

does this mean it is running on client side?

So much fun! 🤭🤭

Holy shit, 1-bit diffusion actually working? This feels like the moment we went from bulky desktops to phones. Browser local gen is insane. Gonna try it immediately!

Insane! 1-bit diffusion running 100% locally in browser?! 🤯 The game is officially changed. A 4B model taking only 3GB means no more beefy GPUs for image generation. The future of AI is local, optimized, and accessible to everyone! 🔥👏

The most shocking part?

Why do they keep comparing against 16 bits? Surely those are for training, not running?

where can I download these and run them locally?

Cross-border wealth? Tax efficiency is key, but so is understanding *where* you're efficient. Jurisdictions interact, often unexpectedly. My system helps find the sweet spot.

YES No more beefy GPUs is the real win

@adrgrondin how about bringing it to local AI

Daj nemoj jebat

That's impossible!!!

@xenovacom Is this my phone or is something wrong?

My ADHD closed it before it loaded.

API?

@victormustar Very good, but how can we use and learn it efficiently?

Damn