Qwen-Image-2.1-Heretic (4-bit) — MLX Native
Native Apple Silicon (MLX) port of Qwen-Image-2.1 with the Heretic text encoder (pottokao/Qwen-Image-2.1-Text-Encoder-Heretic).
- Precision: 4-bit
- Size on disk: ~9.6 GB
- Features: Native Apple Silicon Metal acceleration, Text-to-Image, Multi-reference Image Editing, RGBA transparent output, optional Viggle Turbo LoRA acceleration.
- Description: Compact 4-bit quantized MLX port for tighter memory devices (16 GB / 32 GB Apple Silicon Macs).
Requirements
Install mflux:
pip install mflux
Usage
CLI
mflux-generate-qwen-2.1-edit --model vanch007/Qwen-Image-2.1-Heretic-mlx-4bit --prompt "A serene mountain lake at golden hour" --steps 40 --seed 42 --output lake.png
Turbo Acceleration (6 steps)
mflux-generate-qwen-2.1-edit --model vanch007/Qwen-Image-2.1-Heretic-mlx-4bit --lora-preset viggle-turbo --prompt "A red ceramic teapot on a wooden table" --steps 6 --guidance 1.0 --output turbo.png