Reposted from author: Nunchaku Qwen-Image (Nunchaku quantized version) generates high-quality images from text prompts, advancing the development of complex text rendering. It has been optimized for efficient inference while keeping image quality loss to a minimum.

int4, r128, 4-step Lightning v1.0. The int4 model is suitable for NVIDIA 40-series and earlier GPUs.

CFG: 1.0 Steps: 4-6 steps Sampler: res_multistep Scheduler: beta/sgm_uniform

Generating images via the online WebUI may sometimes produce black images or fail to generate.