MOSS Image learns image editing fast; AI Toolkit adds edit-training support
ostrisai · x · 2026-10-17
AI Toolkit author ostris tested OpenMOSS's MOSS Image and found it learns editing concepts almost instantly: after training on a single "make this person a cyclops" concept, samples immediately showed grids of one-eyed Mona Lisas and portraits. He has added support for training it as an edit model in AI Toolkit (MOSS-Image Base [Edit Training] architecture with Low VRAM, KV Cache options) and is now training a style-reference LoRA.
More from Multimodal
- 'What if the first humanoid at ADE just wanted to feel something?' — slopcore singularity arrives — Yamapama · 2026-10-17
- YouTube Music adds AI labels separating fully, partly, and non-gen-AI tracks — lmoroney · 2026-10-17
- AI influencer workflow: Codex + CapCut Web takes one character from image to video campaign — alifcoder · 2026-10-17
- Pika partners with OpenAI to embed its full video toolset inside ChatGPT — HeyToha · 2026-10-17
- Gemini 4 digs top 5 Grace Hopper anecdotes from CHM oral history into a video — DynamicWebPaige · 2026-10-17
- Real-time video labeling demo tracks a basketball shot's full trajectory live — LinusEkenstam · 2026-10-17