Pruna's distilled Qwen-Image hits ZPix: 1024px in 20s on 8GB GPU

SamuelTallet · reddit · 2026-09-27

Developer SamuelTallet shipped Pruna's distilled Qwen-Image 2.1 (8 steps) as a single bundled model in his open-source tool ZPix. On an RTX 3070M 8GB laptop GPU it generates a 1024x1024 image in 20 seconds once warm. The base model is also included; both support LoRAs, with the base handling them better. Available on GitHub.

Original post →

More from Multimodal

Multimodal channel →