Does MOSS-Image negative prompt actually work? ComfyUI tests suggest it may backfire
Ok_Scientist9269 · reddit · 2026-11-17
A Reddit user tested MOSS-Image's negative prompt in ComfyUI with the default workflow (int8/bf16 base checkpoint, Qwen3.5 9B text encoder as mossimage, Prompt Enhancer disabled, 25 steps, CFG 3, negative correctly wired to the sampler). After multiple seeds it appeared not to work at all — and in some cases seemed to be incorporated into the positive prompt, the opposite of intended. The user is asking for others' experiences; no conclusion yet.
More from Multimodal
- AI-generated cyberpunk loner cityscapes show off cinematic composition skills — Kyrannio · 2026-11-17
- Chalkboard scientific diagram prompt produces stunning hand-drawn anatomy art — azed_ai · 2026-11-17
- Fix the seed for reproducible text-to-image: free course demos NVIDIA's open model Sana — Al_Grigor · 2026-11-17
- REA-Video skill lets Claude reverse-engineer video editing styles; rival claims 20x speed at SSIM 0.96 — toolstelegraph · 2026-11-17
- H3 anime-to-photo conversion impresses: 16MP renders, 6-step LCM sampling — Loose-Journalist-555 · 2026-11-17
- Songwriter finishes AI music video from hospital bed with Suno and Runway — Kyrannio · 2026-11-17