ComfyUI_RH_SmartPhotoCrafter

ComfyUI_RH_SmartPhotoCrafter

• RunningHub China
• RunningHub International

RunningHub ComfyUI wrapper for vivoCameraResearch SmartPhotoCrafter.

This plugin keeps the upstream infer.py implementation and wraps it with a Loader + ImageEdit node pair.

Sources And Links

• Upstream repository: https://github.com/vivoCameraResearch/SmartPhotoCrafter
• SmartPhotoCrafter weights: https://huggingface.co/katelyn2333/SmartPhotoCrafter
• Qwen-Image-Edit-2509 model: https://huggingface.co/Qwen/Qwen-Image-Edit-2509
• Paper: https://arxiv.org/abs/2604.19587
• License: https://creativecommons.org/licenses/by-nc-sa/4.0/

Nodes

• RunningHub SmartPhotoCrafter ModelLoader
  • Loads DiT, VLM, VAE, DiT LoRA, and VLM LoRA.
  • Caches the editor object to avoid reloading large models every run.
• RunningHub SmartPhotoCrafter ImageEdit
  • Inputs one IMAGE.
  • Runs SmartPhotoCrafter's automatic quality/aesthetic analysis and image editing pipeline.
  • Outputs the enhanced IMAGE and generated analysis_text.

Model Layout

Place weights under ComfyUI models:

ComfyUI/models/
  SmartPhotoCrafter/
    dit.safetensors
    vlm/
      config.json
      ...
    dit_lora/
      adapter_config.json
      ...
    vlm_lora/
      adapter_config.json
      ...
  Qwen-Image-Edit-2509/
    vae/
      diffusion_pytorch_model.safetensors
  diffusers/
    Qwen-Image-Edit-2509/
      vae/
        diffusion_pytorch_model.safetensors

The loader uses these fixed model paths under ComfyUI/models:

• dit_path: SmartPhotoCrafter/dit.safetensors
• vlm_path: SmartPhotoCrafter/vlm
• vae_path: diffusers/Qwen-Image-Edit-2509/vae/diffusion_pytorch_model.safetensors
• dit_lora_path: SmartPhotoCrafter/dit_lora
• vlm_lora_path: SmartPhotoCrafter/vlm_lora

These paths are intentionally not exposed as node inputs. Put the model files in this layout before running the workflow.

Example preparation commands from the ComfyUI models directory:

cd /workspace/ComfyUI/models
hf download katelyn2333/SmartPhotoCrafter --local-dir SmartPhotoCrafter
hf download Qwen/Qwen-Image-Edit-2509 vae/diffusion_pytorch_model.safetensors --local-dir diffusers/Qwen-Image-Edit-2509

On CUDA, the loader enables DiffSynth disk offload for the Qwen DiT and VAE models. Keep vram_limit_gb at 0 to auto-use total CUDA VRAM minus 0.5 GB, or set a positive value only when you need a stricter cap.

Basic Workflow

LoadImage -> RunningHub SmartPhotoCrafter ImageEdit -> SaveImage
RunningHub SmartPhotoCrafter ModelLoader -> RunningHub SmartPhotoCrafter ImageEdit

Set width and height to 0 to preserve aspect ratio and resize to target_pixels. The final size is rounded to multiples of 16.

An API-format example workflow is provided at:

examples/smartphotocrafter_basic_api.json

Replace replace_with_input_photo.png with an image available to ComfyUI before submitting the workflow.

Notes

Do not install a different PyTorch build over ComfyUI's existing environment. The included requirements.txt intentionally omits torch, torchvision, and CUDA/NVIDIA packages.

The upstream SmartPhotoCrafter project states that its materials are released under the Creative Commons BY-NC-SA 4.0 license. This license includes non-commercial and share-alike restrictions.