https://github.com/ali-vilab/ACE_plus/tree/main/workflow/ComfyUI-ACE_Plus

https://huggingface.co/ali-vilab/ACE_Plus/blob/main/subject/comfyui_subject_lora16.safetensors

Note:

You need to run the ImageAndMaskPreview node first to get the correct image mask area, so the Redux model can reference it properly. You also need to run that same node to get the correct image height, which you’ll then input into the ImageScaleByAspectRatio node.

The color discrepancy at the edges of the mask is a normal color shift that occurs after the image is processed by the sampler. The solution is to use the basic inpainting workflow for correction. Since it's too straightforward, I didn't integrate it into this workflow.