8-step Turbo LoRA for Qwen-Image-2.1: 6.8x faster, editing quality improves
metalfans · reddit · 2026-11-22
chatrlabs released an 8-step Turbo LoRA for Qwen-Image-2.1 Uncensored Omni, cutting the standard workflow from 35 steps / CFG 5.0 to 8 steps / CFG 1.0. It supports both text-to-image and instruction-based editing with the same LoRA file.
Measured speedup: 54.9s → 8.1s per image (6.8x) on two RTX A6000 48 GB GPUs.
Quality (8-step vs 35-step): prompt adherence and general SFW quality within noise; anatomy 69.4→72.9, image editing 78.9→81.6, identity preservation in human editing 55.8→72.1 — three metrics improved. Known flaw: occasional over-sharpened skin or speckled artifacts.
ComfyUI recipe: LoRA strength 1.0, 8 steps, CFG 1.0, Euler sampler, 1 megapixel, Diffusers sigma schedule via ManualSigmas. Attaches to all 224 adapted modules on BF16/FP8/INT8/NVFP4 builds. Weights, code, and benchmarks are open-sourced.
Related event: Turbo LoRA speeds up uncensored Qwen-Image 6.8x in 8 steps(2 posts)→
More from Multimodal
- Rockstar took 13 years to make games look real; Tesla just made real life look like a game — imjustnewatai · 2026-11-22
- ComfyUI Beginner's First Work: AI Character Covers Hwasa's 'Good GoodBye' — dkjeong0815 · 2026-11-22
- Full indie music video made locally on DGX Spark with Wan2GP + MiniMax H3 — Inevitable_Emu2722 · 2026-11-22
- Open-source Claude skill turns one image into a fully disassemblable 3D mechanical model — tinyfool · 2026-11-22
- Indie dev upgrades InstantFamilyPhoto, his first app that ever sold — tristanbob · 2026-11-22
- ComfyUI usage rankings put Google's Nano Banana family on top, ahead of arena picks — cremebait15 · 2026-11-22