Shared ComfyUI base workflow for the MOSS image model gets community love
Ill_Resolve8424 · reddit · 2026-11-06
A Redditor thanks whoever shared what they call the best MOSS workflow. The screenshot shows a ComfyUI graph for MOSS text-to-image: mossimagebaseint8 diffusion model, Qwen3.5-9B bf16 as the CLIP text encoder (mossimage type), flux2 VAE, ER-SDE sampler with simple scheduler (25 steps, cfg 3.0), a resolution selector at 2MP (e.g. 1184×1776), and a SkipLayerGuidanceDiT node skipping block 22 / single layers 22, 24. Useful as a base template for getting started with MOSS.
Related event: Two-Stage ComfyUI Workflow Unlocks MOSS Image Model(2 posts)→
More from Multimodal
- TailSteer fights drift in multi-turn image editing, model and demo released — Illustrious_Row_9971 · 2026-11-06
- MOSS-image responds better to tag prompts than natural language, users report — VasaFromParadise · 2026-11-06
- 3 Years and Tens of Thousands of Lines of Code vs. One Prompt: AI Scene Generation's Leap — liliyu_lili · 2026-11-06
- New Midjourney --sref style code turns any prompt into flowing brush-calligraphy art — michaelrabone · 2026-11-06
- VideoColorMatch auto-fixes color drift in V2V generations from WAN VACE and Seedance — pftq · 2026-11-06
- 2D-to-3D character design: a flat warrior sketch turned photoreal 3D render — azed_ai · 2026-11-06