Moss turbo without the blur: stacking two turbo LoRAs for realistic photos
kemb0 · reddit · 2026-10-25
Redditor kemb0 shares a ComfyUI setup that removes the characteristic turbo blur of Moss while keeping subject variety at low step counts. Key trick: stack BOTH of Lodestone's turbo LoRAs (newer at strength 1.0, older at 0.5) plus a smartphone-snapshot realism LoRA at 0.5 — each alone hurts diversity or sharpness, but combined they work. Runs 12 steps, 1.5 MP, Euler, Simple, no upscaling, with Solordz Moss Base Int8 checkpoint. Full negative/positive prompts included, plus a recommendation for KJ's Moss Image Prompt Builder node for Ideogram-4-style regional prompting.
More from Multimodal
- Dev vibecodes open-source suite: guided LoRA trainer, dataset captioner, image tournament tool — Speedyrulz · 2026-10-26
- 34MB Kodachrome LoRA for MOSS Image trained on 551 4x5 film scans, one-shot generations — aoleg77 · 2026-10-26
- AI-Generated Fashion Illustrations Show Off Stylized Haute Couture Prompts — azed_ai · 2026-10-26
- AI-generated dark fantasy series 'Buzubuzu' releases 'Daughter of the Sea' episode — Fun_Dragonfly6967 · 2026-10-26
- Melody Protocol releases AI-generated MV 'Himitsu no Love Potion' — Infinite_Print390 · 2026-10-26
- AI-generated anime short 'Rham vs ETHER' shows striking mecha flight sequence — Key_Layer6250 · 2026-10-26