Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > Comfyui-Moss-Image-T8

ComfyUI Extension: Comfyui-Moss-Image-T8

Repo Name

Comfyui-Moss-Image-T8

Author
t8mars (Account age: 0 days)
Nodes
View all nodes(1)
Latest Updated
2026-10-26
Github Stars
0.0K

How to Install Comfyui-Moss-Image-T8

Install this extension via the ComfyUI Manager by searching for Comfyui-Moss-Image-T8
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter Comfyui-Moss-Image-T8 in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

Comfyui-Moss-Image-T8 Description

Comfyui-Moss-Image-T8 brings MOSS-Image Base and SFT image generation to ComfyUI through visual nodes for prompt encoding, sampling, and VAE decoding, with adjustable settings and shared components.

Comfyui-Moss-Image-T8 Introduction

Comfyui-Moss-Image-T8 lets you generate images with MOSS-Image Base or MOSS-Image SFT inside ComfyUI. It brings the official MOSS-Image inference pipeline into a visual workflow, so you can connect nodes, write a prompt, adjust generation settings, and view the finished image on the canvas.

The extension provides separate loaders for the image-generation model, text encoder, and VAE. Base and SFT use the same text encoder and VAE, so you can switch between the two image-generation models without maintaining separate copies of those shared components. The model files preserve the original tensor data and precision, and the inference pipeline is pinned to the official implementation.

If you are new to the extension, start with the SFT example workflow. You can drag its JSON file onto the ComfyUI canvas. You can also drag in the SFT example image, which has its workflow embedded.

Example MOSS-Image SFT output at 1024 × 1024 and 50 steps

How Comfyui-Moss-Image-T8 Works

Think of the workflow as a small production line:

  1. The text encoder reads your prompt. It turns your words into information the image model can use.
  2. The DiT generates a latent image. This is an intermediate representation of the picture, not yet a regular image you can view or save.
  3. The VAE decodes the latent. It converts that intermediate result into a standard ComfyUI IMAGE.

The extension includes six dedicated nodes for loading the three components, encoding text, sampling with the official pipeline, and decoding the result. Its packed latent output uses the MOSS_IMAGE_LATENT type and should connect to this package’s decoder.

The supplied workflows are already set up with the matching model filenames. Both the SFT workflow and the Base workflow are available as JSON files or as generated PNGs with embedded workflows. API workflow files named *-api.json are also included.

The example settings are 1024 × 1024 pixels, 50 steps, CFG 3, shift 5, seed 42, and an empty negative prompt. These are starting settings, not requirements for every image. Width and height must each be positive multiples of 16.

Comfyui-Moss-Image-T8 Features

  • Two model choices: Choose Base or SFT in the supplied workflows. Both use the shared Qwen3.5 text encoder and FLUX.2 VAE.
  • Editable image settings: Change the prompt and sampling settings in ComfyUI. For example, changing the seed gives you a way to explore a different result, while keeping it fixed helps you compare other setting changes more consistently.
  • Ready-to-use visual workflows: Drag a Base JSON workflow, Base PNG workflow, SFT JSON workflow, or SFT PNG workflow onto the canvas. The PNGs are generated images that also contain their complete workflows.
  • Separate model stages: Text encoding, sampling, and decoding use separate components. The extension releases models between these stages to manage memory.
  • Automatic device selection: In auto mode, the extension follows ComfyUI’s device policy and reclaims its cached models before checking available memory. The full text encoder can fall back to the CPU, although that may make generation slower.
  • Optional processing controls: The supplied defaults leave extra SLG and VAE tiling off. The unconditional branch skips layer 22 using zero-based numbering. You can keep the example defaults while learning the workflow, then change one setting at a time to compare results.

The text encoder and DiT use BF16, while VAE decoding uses FP32. The author reports a successful test on Windows with a 24 GB RTX 5090 Laptop GPU; that test is not a stated minimum VRAM requirement.

Comfyui-Moss-Image-T8 Models

ComponentWhat it doesWhen you use it
MOSS-Image BaseProvides one of the two image-generation model variants.Select the Base workflow when you want to generate with Base.
MOSS-Image SFTProvides the other image-generation model variant.Select the SFT workflow when you want to generate with SFT.
Qwen3.5 text encoderInterprets your prompt for the image model.Required by both Base and SFT; you do not need a separate copy for each variant.
FLUX.2 VAEDecodes the generated latent into a viewable image.Required by both Base and SFT.

The ComfyUI-ready model files include separate DiT files for Base and SFT, plus the shared components. Each DiT file is approximately 10.16 GB, the shared text encoder is 16.79 GB, and the VAE is 0.25 GB. Keeping both variants and the shared files takes about 37.4 GB. Each weight file must remain beside its corresponding .assets directory.

The provided context does not describe a specific visual-style or speed comparison between Base and SFT. The most reliable way to choose for your work is to run both supplied workflows with the same prompt and compare the images yourself.

Troubleshooting Comfyui-Moss-Image-T8

A workflow opens, but a loader cannot find a model.

  1. Check that the model files are under the appropriate ComfyUI/models/diffusion_models, text_encoders, and vae directories.
  2. Check that the filenames match those selected in the example workflow.
  3. Make sure each weight file still has its matching .assets directory beside it, including its manifest.json.

A text-encoder loader rejects the tokenizer files.

  1. Open the text encoder’s .assets/tokenizer/ directory.
  2. Confirm it contains the three shipped files: tokenizer.json, tokenizer_config.json, and chat_template.jinja.
  3. Remove any extra files or subdirectories from that tokenizer directory. The loaders reject extras.

An image size is rejected.

  1. Check that both width and height are greater than zero.
  2. Set each dimension to a multiple of 16. The example size, 1024 × 1024, meets this requirement.

The generated latent will not connect to a decoder.

Use this extension’s VAE decoder for its MOSS_IMAGE_LATENT output. After decoding, the result is a standard ComfyUI IMAGE.

Generation is slower than expected.

In auto mode, the full text encoder may run on the CPU when needed, which is slower. The documented 24 GB GPU configuration is a successful test, not a promise that every stage will run on the GPU or a minimum-memory specification.

Frequently Asked Questions

Can I open a workflow from an example image?
Yes. Drag the supplied Base or SFT PNG onto the ComfyUI canvas. Each PNG is a generated image with its complete workflow embedded.

Do Base and SFT need separate text encoders and VAEs?
No. They share the full Qwen3.5 text encoder and FLUX.2 VAE.

Is an empty negative prompt allowed?
Yes. The supplied workflows use an empty negative prompt by default.

Has the author tested the supplied workflows?
The author reports successful 1024 × 1024, 50-step generation using ComfyUI’s canvas Run button for both workflows, along with JSON and PNG import checks. Further details are in the validation notes.

Learn More about Comfyui-Moss-Image-T8

Comfyui-Moss-Image-T8 Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Comfyui-Moss-Image-T8 detailed guide | ComfyUI