First impressions of MOSS-Image: 8-step sampling with surprisingly good anatomy
tricck3zz · reddit · 2026-10-17
First impressions of the MOSS-Image base model with turbo LoRA (8 steps, 1 CFG, euler/simple sampler): versatile output across styles for a small model, with anatomy notably better than many peers. Sample images show a vaporwave fashion portrait.
More from Multimodal
- Ultra-realistic AI puppy alarm-clock photo, full Midjourney prompt shared — tisch_eins · 2026-10-17
- Sigma schedules instead of step counts: 12/14-step sigmas markedly improve Qwen Image Turbo outputs — Sad_Berry_4621 · 2026-10-17
- Indie creator makes K-Webtoon drama with MiniMax H3 and ComfyUI timeline plugin — CirceWebtoonDrama · 2026-10-17
- MOSS Image SFT 8-step turbo LoRA test: 1MP images in 5 seconds on a 4070 Super — Ant_6431 · 2026-10-17
- Open-source local ElevenLabs alternative hits 19.4K stars, dubbing in 646 languages — art_zucker · 2026-10-17
- H3 Character Swap LoRA V1.5 ships with better identity preservation and a free workflow — linoy_tsaban · 2026-10-17