MOSS-Image ecosystem roundup: 314-style catalog, 2-step turbo LoRA, and more

optimisticalish · reddit · 2026-11-01

A community roundup of MOSS-Image resources as of November 1:

• Style catalog fully online: the MOSS-Image Character Catalog now offers 314 styles with their full prompts, free to browse (Hugging Face Space + GitHub), replacing the old limited demo plus 2GB local download.
• MOSS detailer: a slider for adding detail without (apparently) changing style; no trigger needed.
• New style LoRAs: a double-exposure style (trigger ral-dubl, good for movie posters and Steranko-style splash pages) and an isometric miniature style (trigger ral-cubiq; adding "a closeup view of a small isometric model" makes minis more compact).
• Training guide: detailed settings and troubleshooting for training a character LoRA on a single 3060 12GB.
• First hair concept LoRA: a male flat-top hair concept LoRA.
• 2-step turbo LoRA: MOSS-Image-IDM-2-Steps for the bf16 SFT checkpoint, requiring a special loader node; the rapid-moss custom node is recommended since stock nodes can't express the data-space sampler.

Links to the previous week's daily roundups are included.

Original post →

More from Multimodal

• Reddit's viral AI short 'Gnomes' Balloon' charms with a gnome and glowing jack-o'-lantern — love1008 · 2026-11-01
• Shizuku AI unveils in-house singing lip sync tech, beating uLipSync in quality — bdsqlsz · 2026-11-01
• Music video VFX ask: swapping cars for elephants while keeping lip sync intact — gizy843 · 2026-11-01
• inclusionAI Open-Sources IDM: 2-Step Distillation Beats 8-Step Z-Image-Turbo on Chinese Text — bdsqlsz · 2026-11-01
• Midjourney v8.2 Day of the Dead portrait demo with full prompt and parameters — michaelrabone · 2026-11-01
• Teacher's AI-generated fake Pollock fools no one: students spot the difference — Linahuaa · 2026-11-01

Multimodal channel →