Kromoss 8-Step DMD2 ComfyUI Workflow (INT8, 16 GB)
==================================================

Article: https://freakyjolly.com/kromoss-comfyui-8-step-workflow-16gb/

What's inside:
  - Kromoss_8Step_DMD2_16GB.json: ComfyUI workflow (UI format). Drag it onto the ComfyUI canvas or use Workflow > Open.

Required models (download each file and put it in the folder shown):
  - 8-steps-dmd2-lora.safetensors
      folder: ComfyUI/models/loras/
      download: https://huggingface.co/lodestones/Kromoss/resolve/main/8-steps-dmd2-lora.safetensors
  - flux2-vae_small.safetensors
      folder: ComfyUI/models/vae/
      download: https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/vae/flux2-vae_small.safetensors
  - kromoss_int8_convrot.safetensors
      folder: ComfyUI/models/diffusion_models/
      download: https://huggingface.co/lodestones/Kromoss/resolve/main/diffusion_models/kromoss_int8_convrot.safetensors
  - qwen3.5_9b_w4a8_gptq.safetensors
      folder: ComfyUI/models/text_encoders/
      download: https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_w4a8_gptq.safetensors

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

How to use:
  1. Update ComfyUI to v0.42.0 or newer (all nodes are core) and restart.
  2. Download the four files into the folders listed.
  3. Drag Kromoss_8Step_DMD2_16GB.json onto the canvas.
  4. Type your prompt in the "moss-image Positive" node and queue (8 steps, CFG 1, euler / simple, 1024 x 1024).

Notes:
  - If ModelAttentionBackend ("comfy kitchen attention") errors on your build, bypass it.
  - Use one turbo LoRA at a time (do not stack DMD and DMD2).

Credits and licence:
  - Based on the official kromoss_simple_workflow.json from lodestones/Kromoss (Apache 2.0): https://huggingface.co/lodestones/Kromoss
  - Changed: UNETLoader points at kromoss_int8_convrot.safetensors (the saved name kromoss_bf16_11000.safetensors is not in the repo), VAELoader points at flux2-vae_small.safetensors as listed on the model card, and the Note now says the LoRA goes in models/loras/ (the original said models/vae/). Nodes, links and sampler settings are unchanged.

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