logo
RunComfy

AI model workflow

MOSS-Image ComfyUI Workflows

Create images from prompts or plan text and object placement with bounding boxes. Explore MOSS-Image ComfyUI workflows for SFT, Base, and Rapid-MOSS 4-step generation. Run an example on RunComfy Cloud or download its workflow JSON for your own setup.

5 workflows

Run nowView all workflows

Featured workflow

MOSS-Image Rapid-MOSS SFT 4-Step | Fast Text2Image

The workflows

5 MOSS-Image ComfyUI Workflows

Choose a MOSS-Image ComfyUI workflow by checkpoint and input: a text prompt, a bounding-box layout, or the Rapid-MOSS four-step setup. Open any example for its instructions and JSON download. Run it on RunComfy Cloud, download the workflow JSON, or deploy supported workflows as APIs.

5 of 5 workflows shown

About MOSS-Image

What can MOSS-Image do in ComfyUI?

MOSS-Image is an image-generation model family from OpenMOSS. It works with written scene descriptions and structured layouts, making it useful for illustrated scenes, portraits, and designs that combine imagery with text.

Start with the SFT text-to-image example for everyday prompting. SFT is fine-tuned from MOSS-Image Base; both have dedicated layout examples on this page. Choose one of those when you want to draw regions for subjects and headlines before generating.

Rapid-MOSS adds a community four-step adapter to MOSS-Image SFT for quicker iteration. Its adapter and sampling setup work together, so use the dedicated example instead of reducing the standard workflow to four steps. Base, SFT, and Rapid-MOSS are distinct starting points; keep the components specified by each workflow.

References and model resources: OpenMOSS MOSS-Image project, Official MOSS-Image ComfyUI guide, Comfy-Org MOSS-Image files, Rapid-MOSS adapter instructions, Explore MOSS-Image models on Hugging Face, Explore MOSS-Image models on Civitai.

Capabilities

Explore MOSS-Image capabilities in ComfyUI

Start from a written idea, or specify where text and subjects belong before generating. SFT and Base examples support both approaches.

Text-to-image generation

Describe a scene and generate an image with SFT. Use prompt upsampling to expand a short idea, then adjust the seed and image size.

Run now

Layout-guided generation

Draw bounding boxes for text and objects, describe each region, and use the layout to guide the composition.

Run now

MOSS-Image on RunComfy

Run MOSS-Image without assembling a local setup

Open a prepared ComfyUI environment with the model files and nodes used by your selected example. Change prompts, layouts, and sampling controls in the graph, then download the JSON when you want to reuse the workflow in a compatible setup.

Run a MOSS-Image workflow now

  • Editable prompts and layouts

    Inspect expanded prompts or adjust text and object boxes in the example you choose.

  • Prepared model components

    Start with the checkpoint, text encoder, decoder, and nodes configured for that workflow.

  • Cloud GPU choice

    Select hardware for your model format, image dimensions, and additional processing stages.

  • Reusable workflow JSON

    Download the graph to inspect or customize it in your own compatible ComfyUI environment.

Applications

What you can make with MOSS-Image

Develop portraits, illustrated scenes, poster ideas, and product campaign concepts. Choose a layout example when headline and subject placement matter.

Product Campaign Concepts

Arrange a described product, headline, and background into a campaign composition using the Base layout example.

Run now

Frequently asked questions

MOSS-Image in ComfyUI: FAQ

Choose Base or SFT, understand layout and prompt controls, and find the right LoRA training or workflow download path.

Should I use MOSS-Image Base or SFT in ComfyUI?

Start with the SFT text-to-image workflow for general image creation. SFT is fine-tuned from Base, with a focus on more organized layouts and text grouping. Choose the Base example when you want to explore the original checkpoint. Both versions also have layout-guided examples here.

How do MOSS-Image bounding-box workflows control a layout?

They turn boxes and regional descriptions into a structured prompt that guides generation. In the SFT layout workflow, define regions for objects and text, add your wording, and inspect the assembled prompt before generating. Keep the layout canvas and output dimensions aligned. Boxes guide placement; review the result and adjust regions that need more space.

Can MOSS-Image render English and Chinese text in images?

Yes. MOSS-Image supports English and Chinese text rendering. The layout examples let you place headlines and captions alongside image regions. Use short, well-spaced text and inspect spelling and letterforms after generation, especially for small print or dense designs.

When should I turn off MOSS-Image prompt upsampling?

Turn it off when you want to preserve an exact prompt or structured layout. Upsampling expands a short idea into a fuller description; the SFT text-to-image example enables it by default and lets you preview the resulting prompt. The SFT bounding-box example disables it by default so the box-derived layout passes through unchanged.

How do I run MOSS-Image in four steps with Rapid-MOSS?

Use the dedicated Rapid-MOSS SFT 4-step workflow. It combines SFT with the community adapter, its additional assets, and a matching sampler. Lowering the step count in a standard SFT graph does not reproduce that setup. Keep the supplied adapter and sampling components together, and review outputs for your intended style and level of detail.

Should I download MOSS-Image INT8 or BF16, and how much VRAM do I need?

Choose the files required by your selected workflow. Comfy-Org provides INT8 and BF16 variants, with separate text-encoder and VAE files; the format affects memory use. Total VRAM also depends on resolution, offloading, and any extra stages, so checkpoint file size alone is not a reliable minimum. A quantized standard workflow and the Rapid-MOSS adapter setup can have different requirements. Check the detail page before downloading, or use a prepared RunComfy Cloud environment.

Can I train a MOSS-Image LoRA on RunComfy?

Yes. RunComfy Trainer offers MOSS-Image SFT and MOSS-Image Base for text-to-image LoRAs. For a learned editing task, choose SFT Edit Training or Base Edit Training. The edit modes are experimental and use paired reference and target images. Follow the matching training guide for inference requirements; edit LoRAs require the ComfyUI-MOSS-Image-Ostris-Edit nodes.

Where can I download a MOSS-Image ComfyUI workflow JSON?

Use Download JSON on this page or the selected workflow's detail page, then open the file in ComfyUI. The JSON contains the graph, not the model weights or custom nodes. Install the components listed for that example; Rapid-MOSS requires its adapter setup as well as the underlying image model. Run now opens the corresponding prepared cloud workflow.

Related resources

More ways to use MOSS-Image

Train a MOSS-Image LoRA with an SFT or Base checkpoint. Choose a standard text-to-image preset or an experimental Edit Training mode for paired reference and target images.

Train a MOSS-Image SFT Edit LoRA

Teach an editing task with the experimental SFT Edit Training mode using paired reference and target images.

Open SFT Edit Training

Train a MOSS-Image SFT LoRA

Train a subject or style LoRA on MOSS-Image SFT for use with a compatible SFT text-to-image setup.

Open SFT Trainer

Train a MOSS-Image Base LoRA

Train a custom image LoRA on the original MOSS-Image Base checkpoint and evaluate it with the same base model.

Open Base Trainer

Train a MOSS-Image Base Edit LoRA

Use the experimental Base Edit Training mode to learn a reference-guided edit from paired images and instructions.

Open Base Edit Training

Ready to create with MOSS-Image in ComfyUI?

Run now