I have a full guide on using this! Take some time to look at it: https://civarchive.com/articles/38065/succupon-all-in-one-moss-workflow-guide

This is a condensed version of that guide.

Design Philosophy
The goal of this workflow is to be as feature rich as possible through a single pane of glass. This inherently introduces complexity. To counter this, I have rigorously gone through my previous workflows and have cut as many custom nodes as possible. This is more work for me as I generally need to stitch together Comfy Core nodes to get what I want, but it requires less work for the user and also increases stability when node packs get updated.

You will not find sub-graphs here; everything is configurable on a single page.

Basic Configuration
As a basic rule, purple nodes require configuration, grey nodes do not. Positive prompts are green. Negative prompts are red. Documentation and guides are yellow. There are some deviations but this is generally the case in this workflow

The bypasser section has multiple groups that can easily be toggled on or off. By default, the workflow ships in Text2Image mode with upscaling enabled.

Lora loading is handled by Power Lora Loader.

Like usual, you can search Civitai data on the Lora with the following


Prompting
You have full wildcard support via Comfy Impact Pack. Full documentation on how to use this can be found here. https://github.com/ltdrdata/ComfyUI-Impact-Pack

Prompt Generator
This uses the default CLIP Loader as an LLM to process your prompt. I have included a custom system prompt.

It is work noting that the recommended CLIP can censor prompts. I tried to account for this but it is not perfect.

If you want to use a different LLM that is uncensored, you can enable Custom Clip. I recommend the following: https://huggingface.co/ahmed22xa/Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy/resolve/main/Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled.safetensors


You can also use an image to generate a prompt with the following option:


Style Library

This section contains 400+ styles to choose from. Selecting one will append your prompt with natural language to guide it towords the style. You can read more about it here https://github.com/lumenastrum/clio-style-preview


BBOX Prompts

Draw bounding boxes and prompt for each individual box. This is one of my favorite features of MOSS.
The Scene/Background box is not meant to be filled in. The primary Positive Prompt ImpactPackWildcard Processor will feed into this section automatically. This allows you to use wildcards, feed the scene/background through the style library, or feed the prompt through the prompt generator.

Draw your boxes and input a prompt for what should be in each colored box. Not that there are drop downs on each prompt box to modify things like camera angles, subjects, etc.

At the bottom you will see settings for the image size, and camera settings. These settings will override any other conflicting settings. (Example: If your latent size is set to 1024x1024, and your general configuration is set to 2MP, the final image will be 1024x1024.) You can also save presets for common resolutions and settings.

The node will output a prompt and automatically calculate the pixel boundaries for each box in the prompt.

Below in the Upscaler section you can see the resulting image from this setup. The results are quite impressive and only one generation was needed to get them!

Upscalers

This workflow comes with three upscalers; Ultimate SD Upscale, SeedVR2, and RTX Super Resolution.

SeedVR2 and Ultimate SD Upscale refine images that pass through them. Ultimate SD Upscale is a bit more aggressive in doing this, whereas SeedVR2 is more conservative. Both will upscale by 2x and a single upscale will bring the image to 4K resolution. You can also enable both if desired. RTX Super Resolution does not refine the image

I personally recommend SeedVR2 mostly because it is quicker, and MOSS doesn’t need to much refinement after the image is generated.

Other Options

Turbo Mode: This will significantly improve generation time, but you may get less prompt adherence and quality. The workflow ships configured for the lodestone 8 step Turbo LoRA, so there is no need to download a specific Turbo model.


With this option enabled, your CFG, Steps, and Scheduler will be overrode in your generation. Any modification to Turbo mode values should be done here and not in the Global Configuration


SLG: SLG is recommended by the creators of MOSS Image in some scenarios. This deliberately skips one or more transformer layers inside the DiT. The layers that are skipped depend on the art style you are trying to achieve. I have included three Presets, and one Custom Present you can modify to your liking. You simply enable Skip Layer Guidance, and select the preset you want.

Ensure only one is selected.



Post Processing: This is just a simple way to modify colors, levels, and apply various effects. The workflow ships with a slight bloom and increase to color vibrancy preconfigured.


Models
MOSS-Image Base CHOOSE ONE based on your needs and hardware

- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_base_int8_convrot.safetensors

- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_sft_bf16.safetensors

Text_Encoders CHOOSE ONE based on your needs and hardware
- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_int8_convrot.safetensors

- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_w4a8_gptq.safetensors

VAE CHOOSE ONE based on your needs and hardware

- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/vae/flux2-vae.safetensors

- https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/vae/flux2-vae_small.safetensors

Upscale Models

- https://huggingface.co/Kim2091/UltraSharpV2/resolve/main/4x-UltraSharpV2_Lite.safetensors

- SeedVR2 (This downloaded automatically when you run it for the first time.)

LoRA

- Turbo: https://huggingface.co/lodestones/Kromoss/resolve/main/8-steps-dmd2-lora.safetensors

LLM

- https://huggingface.co/ahmed22xa/Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy/resolve/main/Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled.safetensors

Only if you want to use a different LLM for the Prompt Generator

Custom Nodes Used

- rgthree/rgthree-comfy https://github.com/rgthree/rgthree-comfy

- ltdrdata/ComfyUI-Impact-Pack https://github.com/ltdrdata/ComfyUI-Impact-Pack

- numz/ComfyUI-SeedVR2_VideoUpscaler https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler

- ssitu/ComfyUI_UltimateSDUpscale https://github.com/ssitu/ComfyUI_UltimateSDUpscale

- alexopus/ComfyUI-Image-Saver https://github.com/alexopus/ComfyUI-Image-Saver

- ukr8b3g-cmyk/MossImage-BBOX-Prompter https://github.com/ukr8b3g-cmyk/MossImage-BBOX-Prompter

- 0xBeycan/ComfyUI-BCNodes https://github.com/0xBeycan/ComfyUI-BCNodes

- lumenastrum/clio-style-preview https://github.com/lumenastrum/clio-style-preview

- Comfy-Org/ComfyUI_Nvidia_VFX_Nodes https://github.com/Comfy-Org/ComfyUI_Nvidia_VFX_Nodes

- ltdrdata/was-node-suite-comfyui https://github.com/ltdrdata/was-node-suite-comfyui