MOSS Image 5B Base tested: no LoRA, no finetune, straight gens at 50 steps / CFG 3
Unit2209 · reddit · 2026-10-23
A Reddit user posted a collection of straight generations from the MOSS Image 5B Base model (not the SFT variant) — no LoRA, no finetuning. Most images used standard settings: 50 steps, CFG 3, 1.5 megapixels. Raw image files are shared via Google Drive, and the author offers to reply with the specific workflow for any image by number, making results reproducible.
More from Multimodal
- Pika lands inside ChatGPT: one prompt generates a full AI wrestling match — aziz4ai · 2026-10-23
- Six cinematography lighting rules to fix flat AI shots, straight into your prompts — lmoroney · 2026-10-23
- AI Video for Less Than a Movie Ticket: Dreamina at $0.29 Plus CapCut PC Workflow — JaynitMakwana · 2026-10-23
- AI video hook technique: fill the frame with text before any footage appears — aftahi_ai · 2026-10-23
- Design explorations for Vercel's First of Kind season 3 revealed — soleio · 2026-10-23
- Amazon Polly Generative engine upgrade adds 14 voices, first Hindi and Finnish generative voices — AWS ML Blog · 2026-10-23