Three things that cost me an evening installing MOSS-Image in ComfyUI, and none of them was the model

Description and use

Installing MOSS‑Image in ComfyUI requires correct file placement, node pack installation, and variant selection; without these steps the UI runs silently and outputs noise.

Problem

Installing MOSS‑Image in ComfyUI requires correct file placement and matching model components, otherwise the UI runs silently and produces noise instead of an error.

Outcome

Correct file placement, node pack installation, and variant selection enable MOSS‑Image to run smoothly in ComfyUI, revealing its layout control capabilities.

Workflow

• 1. Download MOSS‑Image checkpoint and text encoder files from the official OpenMOSS repo.
• 2. Place the checkpoint in ComfyUI/models/diffusion_models/ and the two text encoder files in ComfyUI/models/text_encoders/.
• 3. If using a separate VAE, place it in ComfyUI/models/vae/.
• 4. Install the required node pack via the manager or clone it into custom_nodes/ and restart ComfyUI.
• 5. Verify the node appears and no "node does not exist" error occurs.
• 6. Choose the correct variant (SFT or Turbo) and set sampler steps accordingly (20–30 for SFT, 4–8 for Turbo).
• 7. Build a minimal text‑to‑image graph (checkpoint, encoder, short prompt, 1024×1024) and confirm a picture is produced.
• 8. Once confirmed, build the layout workflow using bounding boxes or JSON layouts.
• 9. Test layout control by planning arrangement first and describing each region.
• 10. Iterate quickly due to the modest 5B parameter count.

Evidence

The text encoders are separate files, and a missing one gives you garbage instead of an error

Source

If one is missing or mismatched, ComfyUI does not tell you a file is missing. It runs, and produces noise.

Source

Install the node pack from the manager or clone it into custom_nodes/, then restart ComfyUI.

Source

Variant mismatch. There are two: SFT (base) and Turbo (a LoRA, far fewer steps). The sampler settings differ.

Source