Combating MOSS Image blur with a mid-denoise latent sharpening trick
LMLocalizer · reddit · 2026-10-22
A Reddit user shares a simple two-stage workflow to reduce blur in MOSS Image outputs:
- Denoise normally for the first 5 steps;
- Apply LatentOperationSharpen to the latent, then continue denoising.
Tested on the base model with the Kromoss 8-steps-dmd2 LoRA, with two ComfyUI workflows published: one also stacking the extracted MOSS-Image-SFT LoRA during the second stage, and one with latent sharpening only. No training required, and the approach may transfer to LoRA-free setups.
More from Multimodal
- Higgsfield CEO: 50%+ of social content will be AI-generated by end of 2027 — SimplyAnnisa · 2026-10-22
- Midjourney sref combo generates charming oil-paint sports cats — sergeantsref · 2026-10-22
- Custom node pack brings looping H3 video generation to Comfy Cloud — PurzBeats · 2026-10-22
- LiDAR scan + LLM workflow produces dimensionally accurate 3D room models — OpenAIDevs · 2026-10-22
- Team spent $26,500 on short film Nexus 6 to build a Blender workflow that halved costs — NightsRadiant · 2026-10-22
- AI short film recreates New York's infamous 1858 blue milk scandal — hellomyoldfrien · 2026-10-22