MOSS-Image Edit LoRA ComfyUI Workflow (native nodes, single reference)
======================================================================

Article: https://freakyjolly.com/moss-image-edit-lora-comfyui/

What's inside:
  - MOSS_Image_Edit_LoRA_Native.json: ComfyUI workflow (UI format). Drag it onto the canvas.
  - MOSS_Image_Edit_LoRA_Native_api.json: Same graph in API format.

Required models (download each file and put it in the folder shown):
  - moss_image_sft_bf16.safetensors
      folder: ComfyUI/models/diffusion_models/
      download: https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_sft_bf16.safetensors
  - moss_image_edit.safetensors
      folder: ComfyUI/models/loras/moss-image-edit/
      download: https://huggingface.co/chinoll/Moss-Image-Edit/resolve/main/adapter_model.safetensors (rename to moss_image_edit.safetensors)
  - qwen3.5_9b_int8_convrot.safetensors
      folder: ComfyUI/models/text_encoders/
      download: https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_int8_convrot.safetensors
  - flux2-vae.safetensors
      folder: ComfyUI/models/vae/
      download: https://huggingface.co/Comfy-Org/Fibo-2.0/resolve/main/vae/flux2-vae.safetensors

Custom nodes:
  - None. Every node is part of ComfyUI core (update ComfyUI to the latest version).

How to use:
  1. Update ComfyUI to a build newer than v0.42.1 (Text Encode MOSS-Image Edit was added in PR #17296) and restart.
  2. Download the four files into the folders listed and rename adapter_model.safetensors to moss_image_edit.safetensors inside models/loras/moss-image-edit/.
  3. Drag MOSS_Image_Edit_LoRA_Native.json onto the canvas.
  4. Pick your photo in "Image 1", type the edit instruction in the positive Text Encode MOSS-Image Edit node and queue.

Notes:
  - This is Route 2 from the article (core nodes only). Settings follow the node pack defaults the article lists: reference longest edge 512, 30 steps, CFG 4, euler / normal, seed 42 fixed. The output size follows the scaled photo.
  - Shift: ComfyUI configures MOSS-Image with shift 5.0 automatically, so no extra model sampling node is needed.
  - For 2 to 4 references, connect more Load Image nodes to the images inputs of BOTH Text Encode nodes (the core node says to use it for the negative too) and refer to them as image 1, image 2... in the prompt.
  - On 16 GB start ComfyUI with --lowvram and stay at 512 to 768 px while testing. I have not benchmarked this graph.

Model weights are not included. Each model keeps its own licence; check the model page before commercial use.
Workflow shared by FreakyJolly.com.
