mlx FLUX.1 schnell 4bit quantized
by argmaxinc
Open source · 9k downloads · 31 likes
1.9
(31 reviews)ImageAPI & LocalAbout
The mlx FLUX.1 schnell 4-bit quantized model is an optimized and streamlined version of FLUX.1 schnell, specifically designed to run efficiently on Apple devices equipped with M-series chips using the MLX library. It is a diffusion-based image generation model capable of producing high-quality visuals from text prompts in real time or near real time. Its core capabilities include generating realistic, artistic, or stylized images with high fidelity to the provided prompts. The model stands out for its speed and energy efficiency, thanks to 4-bit quantization, which reduces parameter size without significantly compromising output quality. It is particularly well-suited for content creators, mobile or desktop application developers, and users seeking a high-performance, locally run image generation solution.
Documentation
FLUX.1-schnell on DiffusionKit MLX!
Check out the original model!
Check out the DiffusionKit github repository!
![FLUX.1 [schnell] Grid](./flux_on_mac.png)
Note: This checkpoint features 4-bit quantization of the mmdit module using MLX's nn.quantize function with default settings (group_size=64).
Usage
-
Create conda environment
Shell
conda create -n diffusionkit python=3.11 -y
conda activate diffusionkit
pip install diffusionkit
-
Run the cli command
Shell
diffusionkit-cli --prompt "detailed cinematic dof render of a \
detailed MacBook Pro on a wooden desk in a dim room with items \
around, messy dirty room. On the screen are the letters 'FLUX on \
DiffusionKit' glowing softly. High detail hard surface render" \
--model-version argmaxinc/mlx-FLUX.1-schnell-4bit-quantized \
--height 768 \
--width 1360 \
--seed 1001 \
--step 4 \
--output ~/Desktop/flux_on_mac.png
Capabilities & Tags
diffusionkittext-to-imageimage-generationfluxmlxen
Links & Resources