ComfyUI v0.41.3 fixes FLUX 3 Klein launch workflows: prompt upsampler on, 50 steps
Lexius2129 · reddit · 2026-10-18
ComfyUI team member Alexis acknowledges that FLUX 3 Image Klein's day-one example workflows had sub-optimal settings causing bad anatomy, and reworked them with Black Forest Labs in v0.41.3:
- Text-to-Image (base): prompt upsampler on by default, 25→50 steps; distilled: 15→16 steps
- Image Edit: prompt upsampling added for both variants; base edit 25→50 steps
- Upsampler: new 4-bit quant (qwen3.59b, 7GB vs 19GB bf16) with negligible quality loss; BFL recommends always using it
- Renamed model files: base = undistilled (50 steps, CFG), no base = step-distilled (16 steps, CFG 1); bf16/int8/nvfp4 quants available
To update: upgrade to v0.41.3, open the FLUX 3 Klein templates fresh from the Template Library, and download the new upsampler. Anyone who judged the model on launch workflows should rerun it.
Related event: BFL and ComfyUI ship fixes for FLUX 3 Klein launch issues(4 posts)→
More from Multimodal
- Priming Models: one-step generation hits 1.19 FID with 12x less compute, no diffusion teacher — LucaAmb · 2026-10-20
- LILIE: language-informed lossless image encoder improves gFD by 14 points on GPIC — jcniebles · 2026-10-20
- Indie dev builds custom AI tools to control generative output instead of pulling the slot machine — rms80 · 2026-10-20
- AI-Generated Sci-Fi Short 'Alien Invasion Prologue' Wows Reddit With Cinematic Look — theodore_70 · 2026-10-20
- AI-Generated Harry Potter Clip Shows Harry and Hermione's Confession Scene — lightandcolorlab3 · 2026-10-20
- HeyGen Video 1.0 Launches, Ranks #3 on Design Arena and Makes HeyGen #2 Video Lab — HeyGen · 2026-10-20