Why MOSS Image looks better: it deliberately excluded AI images from training data
ostrisai · x · 2026-10-18
ostrisai argues the key to MOSS Image's popularity is that the team intentionally removed AI-generated images from its training data. He sees a clear trend of new image models training on synthetic data and converging on an overly 'ChatGPT look', while MOSS avoids that aesthetic degradation.
More from Multimodal
- Voice demo built with Jev draws "holy crap, that's the missing thing to get rid of keyboards" reaction — schwentker · 2026-10-18
- Video-Index 1.1: agentic video understanding beats feeding 512 fixed frames — thoma_gu · 2026-10-18
- Openspline: open-source realtime talking avatar hits 90+ FPS on an RTX 4090 — Mikasa0xdev · 2026-10-18
- Pixel-art cat curled around a planet, generated by Opus 5.5 — liminal_bardo · 2026-10-18
- One prompt, a 3-minute animated film: Claude Fable 5.5 goes from human civilization to 30 years in the future — FinanceYF5 · 2026-10-18
- Local anime video generation with MiniMax H3 hits style drift problems — Physical-Resort-4104 · 2026-10-18