AI Model Release
Summary
PrismML releases Bonsai Image 4B, a compact diffusion model for local inference on iPhones and Macs. It features 1-bit and ternary variants that drastically reduce model size while maintaining competitive image quality and speed.
Highlights
- Enables on-device generation on iPhone 17 Pro Max and Mac M4 Pro.
- 1-bit variant reduces transformer footprint to 0.93 GB, an 8.3x reduction.
- Ternary variant preserves 95% performance with 1.21 GB size.
- Built on FLUX.2 Klein 4B architecture with binary and ternary weights.
- Inference takes 9.4 seconds on iPhone and 6 seconds on Mac M4 Pro.
auto-generated
Context
Audience
AI Researchers and Mobile Developers
DomainMachine Learning
Formattechnical announcement
Accessopen weights
Topics
Related
FLUX.2 Klein 4BModel QuantizationDiffusion TransformersApple Silicon OptimizationOn-Device AI