MOSS Image in ComfyUI: Put Each Model File in the Right Place
Understand Comfy-Org’s MOSS packaging, its three model components, and a repeatable first-generation workflow.
AI SearchA downloaded image model will not run if its workflow cannot find the matching text encoder and decoder. The Comfy-Org MOSS Image repository addresses that packaging problem by providing model files for ComfyUI and documenting their destinations.
Three components to match
The Comfy-Org model card identifies its files as repackaged releases and links to OpenMOSS’s original Base and SFT models. It lists the following arrangement:
| Component | ComfyUI folder | Role |
|---|---|---|
| MOSS diffusion model | models/diffusion_models/ | Generates the image representation. |
| Qwen3.5 text encoder | models/text_encoders/ | Converts the prompt into conditioning. |
| FLUX.2 VAE | models/vae/ | Decodes the representation into pixels. |
Choose filenames explicitly supported by your workflow. Alternative precision formats are different loading choices, not extra models that must all be downloaded.
Establish one baseline image
Start with a supported example workflow, select the files in its loader nodes and make one generation before adding adapters. Record the ComfyUI version, model filenames, prompt, dimensions, seed and sampler settings. If a node is unavailable, follow the current integration instructions rather than replacing it with a similarly named node.
The original SFT card documents natural-language, tag and JSON-layout conditioning. For an initial test, describe a single object and its background. Introduce text placement and multiple objects only after the basic path works.
What a successful load proves
One output confirms that the chosen files can execute together on your installation. It does not establish a minimum GPU requirement or a ranking against other generators. Inspect actual memory consumption and retain the workflow before trying quantized weights or the separate acceleration adapter. The repository labels the package Apache-2.0; review the relevant upstream components as well when redistributing a complete setup.
Model settings and downloadable workflow
The upstream SFT card specifies the following baseline. These are publisher defaults, not measured recommendations for every GPU.
| Setting | Documented default |
|---|---|
| Sampling | Flow-matching Euler |
| Inference steps | 50 |
| CFG | 3 |
| Shift | 5 |
| Negative prompt | Empty |
| Layer skipping | Negative branch, index 22, counted from zero |
The official example workflow JSON is an actual workflow asset to inspect and import using the upstream instructions. It is not a model-weight download. The user guide explains prompt formats and experimental layouts. The supplied layout illustration below is publisher evidence, not an independently reproduced output.
Source image: OpenMOSS Team. Publisher-provided example; Qentrah has not reproduced this result.
Source video and image gallery
The source tutorial covers loading MOSS in ComfyUI and layout control. The chapter links are taken from its published description.
ComfyUI setup chapter, 03:01. The embedded player above opens the source video; timestamps here open its relevant chapter. Video titles and chapter labels were checked against published metadata, not by claiming to have watched the footage.
Source image: OpenMOSS Team. Repository or publisher artwork, retained here alongside the video.
Source image: OpenMOSS Team. The publisher uses this example to illustrate SFT text grouping; it is not an independent test.
Sources and discovery
This topic comes from link 2 in the description of AI Search’s MOSS Image video. The description was inspected; this article is not a transcript or a claim that Qentrah watched the footage.
Original description link · Reference source. Reviewed on 20 October 2026. Qentrah has not independently benchmarked or tested the products discussed.
