Back to blog

MOSS-Image ComfyUI errors: red nodes, wrong folders, SLG

mossimage.dev

Three ComfyUI setup failures with a documented cause: a template paired with the wrong checkpoint, files in the wrong folder, and an SLG message.

MOSS-Image ComfyUI errors: red nodes, wrong folders, SLG

When MOSS-Image fails in ComfyUI, check three documented setup points first: whether the template and checkpoint match, whether each model file is in the right folder and the text encoder uses CLIPLoader type moss_image, and whether you are seeing the unconfirmed SLG validation report from a user discussion. The first two have documented setup guidance; the SLG post does not establish a cause or fix.

TL;DR

  • SFT templates use the SFT checkpoint; the Base checkpoint belongs with the Base setup.
  • Put the diffusion model, text encoder and VAE in their documented folders.
  • The text encoder for the four text-to-image templates and the edit template needs CLIPLoader type moss_image.
  • A user reported an SLG validation message, but nobody in that thread confirmed its cause.
What you see What to check What the source actually says
A template paired with the other checkpoint The templates load moss_image_sft_int8_convrot.safetensors; Base is moss_image_base_int8_convrot.safetensors with its own settings A setup check, not a list of every mismatch symptom
Red nodes Diffusion model, text encoder and VAE each in its documented folder, text encoder on CLIPLoader type moss_image Not every red node comes from these causes
“SLG scale muss ≥ 0 sein …” Hugging Face discussion #10 A user report; nobody in the thread confirmed a cause or a fix

Does the selected template match the checkpoint?

The ComfyUI guide says the templates load the instruction-tuned SFT checkpoint, moss_image_sft_int8_convrot.safetensors. The other documented checkpoint is moss_image_base_int8_convrot.safetensors. The guide gives distinct Base settings: 50 steps, CFG 3 and the euler sampler; the templates use 25 steps, CFG 2.5 and the er_sde sampler. Check which workflow you opened before swapping a checkpoint. The source does not describe every possible result of a mismatch, so this is a setup check, not a diagnosis for an unspecified error.

Are the model files in the right folders and loader?

The documented folder layout separates three types of files:

  • Diffusion model: models/diffusion_models/
  • Text encoder: models/text_encoders/
  • VAE: models/vae/

The named SFT and Base files are diffusion models. qwen3.5_9b_w4a8_gptq.safetensors is the text encoder used by the four text-to-image templates and the edit template; load it with CLIPLoader type moss_image. flux2-vae_small.safetensors is the VAE. If nodes appear red, first compare the folder and loader choices with those documented names and paths. The docs support these as setup checks; they do not establish that every red node comes from one of these causes.

For the full template walkthrough, see the ComfyUI MOSS-Image guide. The MOSS Image tool page is the browser option if you would rather not manage local ComfyUI model files for this workflow.

What does the SLG validation report tell us?

In Hugging Face discussion #10, a user reported the validation message: “SLG scale muss ≥ 0 sein und 0 ≤ SLG start ≤ SLG stop ≤ 1” and said they could not find the fields to change. Nobody in the thread confirmed a cause. That makes it an unconfirmed user report, not a documented MOSS-Image failure mode.

The message itself expresses constraints on SLG scale, start and stop. The discussion does not say which template or node produced it, why the fields were unavailable, or which setting fixes it. Do not treat a guessed setting as a verified solution. Compare your workflow against the documented template and consult the ComfyUI guide before changing values; the supplied evidence does not support a more specific repair.

The two example images below use the same words in a flat poster and in a workshop photo scene. They are prompt examples for the browser studio and do not demonstrate a fix for the reported validation message.

The bicycle repair class, workshop and schedule in these sample images are fictional.

Bicycle repair class poster with a flat bicycle frame illustration

Make one like this Open this image on MOSS Image

Made in prompt mode with this exact prompt:

Bicycle repair class poster, flat vector illustration of a bicycle frame and a chain tool on a plain sand background, rust orange and dark navy type, wide clear margins. Text, render exactly and add nothing else: headline 'BIKE REPAIR'; line 'THURS 7 PM'; line 'BRING YOUR OWN CHAIN'; footer 'WORKSHOP 2'.

Photo-style bicycle repair poster over a tool bench

Make one like this Open this image on MOSS Image

Made in prompt mode with this exact prompt:

Photo of a dim workshop bench with bicycle tools hanging on a pegboard, warm lamp light from above, calm empty wall in the upper third for the type. Text, render exactly and add nothing else: headline 'BIKE REPAIR'; line 'THURS 7 PM'; line 'BRING YOUR OWN CHAIN'; footer 'WORKSHOP 2'.

Which of these problems applies to a hosted studio?

The first two checks concern local ComfyUI templates, model folders and loader configuration. The SLG message is also reported as a ComfyUI validation message. The browser studio runs the same open model with a visual box editor, but its interface does not ask you to place these named files into local model folders or select a local ComfyUI template. Those particular setup steps are therefore not things to debug in the hosted workflow. That does not mean hosted generation cannot have other issues; the supplied sources do not compare all failure types.

MOSS Image Studio is an independent hosted studio for the open MOSS-Image model. Not affiliated with OpenMOSS.

What are the limits of these troubleshooting notes?

The file names, folders and template settings come from the ComfyUI guide. The SLG message comes from one user discussion and has no confirmed cause in that thread. Red nodes are a symptom mentioned in the topic, but the provided sources do not document a red-node error string or prove a universal cause. Use the documented checks to narrow the setup, and do not present an unconfirmed report as a fix.

FAQ

Which checkpoint do the SFT templates load?

They load moss_image_sft_int8_convrot.safetensors. The Base checkpoint is moss_image_base_int8_convrot.safetensors and has separate settings in the guide.

What CLIPLoader type does MOSS-Image need?

The text encoder for the four text-to-image templates and the edit template is loaded with type moss_image.

Is the SLG validation message a confirmed bug?

No cause was confirmed in discussion #10. It is a user report, so the available evidence cannot establish a bug or a specific fix.

Sources

Try it with your own words

Open the Studio, write a brief or lay out boxes on the canvas, and generate with MOSS-Image.