MOSS image test: 12 distinct characters in one photo with no bleeds or duplicates
Loose-Journalist-555 · reddit · 2026-10-25
A Reddit user tested MOSS's higher token limits by generating a single photo containing 12 different people, using a 1.3k-token prompt with 4-step turbo LoRA on MOSS SFT at 4MP with Euler/simple sampling. Each character's appearance, outfits, and props held with no attribute bleeds and no duplicates — an impressive multi-character consistency result.
More from Multimodal
- Mystery 'Iron Goddess' video model appears on LM Arena, believed from a major Chinese lab — koltregaskes · 2026-10-25
- Dreamlike blue-and-orange AI portraits evoke Harry Potter painting style — azed_ai · 2026-10-25
- Asking Savi to generate every train: the train kid has been digitized — TAbrodi · 2026-10-25
- Third-party providers list uncensored Seedream 5.0 Pro that ByteDance's official API refuses — lizaeve · 2026-10-25
- Midjourney prompt share: getting candid, unretouched morning-light portraits — tisch_eins · 2026-10-25
- Developer one-shots a promo short video with Claude Opus — FinanceYF5 · 2026-10-25