Uncensored Qwen-Image Omni gets Turbo LoRA: 6.8x faster at 8 steps
metalfans · reddit · 2026-11-22
After Reddit users pushed ChatR Labs' Qwen-Image-2.1 Uncensored Omni to #8 trending among 3M+ Hugging Face models, the team released a Turbo LoRA that cuts generation from 54.9s to 8.1s per image.
Key facts
- Runs at 8 steps with CFG 1.0 (vs 35 steps, CFG 5.0), a 6.8x speedup
- Supports both text-to-image and instruction-based editing at 1MP
- Works with ComfyUI and Diffusers; available in GGUF, BF16, FP8, NF4, INT8 ConvRot and NVFP4 builds
Benchmarks (Turbo 8-step vs base 35-step): prompt adherence (89.1 vs 89.7) and general SFW quality (79.0 vs 79.6) are within statistical noise, while anatomy (72.9 vs 69.4), editing quality (81.6 vs 78.9) and identity preservation in human editing (72.1 vs 55.8) improved.
Recommended settings: LoRA strength 1.0, 8 steps, CFG 1.0, no negative prompt. The release includes no prompt filter or output safety checker. Their free Playground now serves 10,000+ images daily, and the team is soliciting stars, testing, GPU resources and sponsorships.
Related event: Turbo LoRA speeds up uncensored Qwen-Image 6.8x in 8 steps(2 posts)→
More from Multimodal
- Indie dev shrinks open-source TTS model Kokoro from 320MB to 40MB with voice intact — kingslyj · 2026-11-22
- 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