BONSAI IMAGE

BONSAI IMAGE was created by researchers from Caltech, and came out on May 26, 2026:

• Load from HERE
• (Hugging Space): prism-ml/Bonsai-Image-Demo

• (YouTube): Self-improving AI, Opus 4.8, Nvidia bangers, game-ready 3D models, juggling robots: AI NEWS
• (YouTube): How to Generate Image LOCALLY with Bonsai Image 4B (It's Only 0.93GB)

BONSAI is a small AI text-to-image that will run on my laptop (and on a smart phone).  It does seem to download a few gigs of the model though:

Images generated with BONSAI

The images are not great, but are about as good as running Stable Diffusion in Automatic 1111 locally on a computer with VRAM.

JAVELINAS in the COFFEE HOUSE

 

I prompted: "Javelinas drinking coffee at a coffee house" in BONSAI and generated the image below.  It's not great, but it worked, and quickly:

Bonsai prompt:

"Javelinas drinking coffee at a coffee house"

THEN I uploaded the BONSAI image to both Midjourney and KREA -- using the same prompt -- to further evolve the image with AI:

Midjourney

Midjourney prompt:

"Javelinas drinking coffee at a coffee house"

KREA

KREA prompt (with an AI model based on my drawings):

"Javelinas drinking coffee at a coffee house"

THEN I mixed the two images above in Midjourney and KREA, and animated the result in Midjourney, just for fun:

Midjourney

KREA

Animated in Midjourney

POTENTIAL

It is a nice tool to use on my laptop.  However perhaps it could be a great tool if I could train the BONSAI model on my drawings.  Or even train BONSAI on some simple schematic-like images, as a way to pre-process an image before uploaded it to Midjourney or a more sophisticated program.  

It would be fantastic for preprocessing!  I would like to upload one of my drawing images into BONSAI, and then alter it with a simple model trained on my drawings or other schematics.