ComfyUI workflow

Background

My spouse works for a small online store selling spare parts globally. They take many product photos and update them regularly on product description pages. Previously, this image processing was time-consuming. Because the photos are taken with a phone, the process involved removing the background or extracting the items, composing the items against a pure white background, or replicating items multiple times in a single image to show stacks, adding text labels, and finally resizing the images.

Last year, I used BRIA-RMBG-2.0 with RMBG-2-Studio to help remove the background and compose the object with a white background, but there was still much work to do. Until recently, I found that BRIA-RMBG-2.0 has a ComfyUI node. I am planning to use this node along with other nodes to fully streamline the whole process.

Setup Proceedure

• install ComfyUI, follow the guides on the page to install it on our home Linux workstaion with nvidia 1070Ti GPU
  • the Linux box is running debian 12 with cuda 12.4
  • clone the repository
  • create python virtual environment python3.11 -m venv venv
  • activate the virtual environment source venv/bin/activate
  • install pytorch packages pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
  • install other requirments pip install -r requirements.txt
• install ComfyUI Manager to manage the ComfyUI
• using ComfyUI Manager to install custom nodes
  • ComfyUI-RMBG, various nodes inlcuding
    • Remove Background (RMBG): used to remove the background
    • BiRefNet Remove Background (RMBG): also used to remove the background
    • Segmentation (RMBG): extract the object if the origianl image has more than one objects
    • Image Preview (RMBG) 🖼️
    • Image Stitch (RMBG) 🖼️: arranges the items and output as an images
    • Image Combiner (RMBG) 🖼️: used to combine the image object with pure white background image
    • Crop To Object (RMBG) 🖼️🎭: used to crop the image to its object size, same as comfy-node-crop-image-by-mask
  • comfy-node-crop-image-by-mask: crop the image to its object size
  • TextOverlay Node for ComfyUI: add text on the images, support alignments
  • ComfyUI_LayerStyle: it has node Brightness & Contrast node as BrightnessContrastV2 to change the image Brightness and contrast
  • ComfyUI_rotate_image: rotate image
  • KayTool: has node 𝙆 Image Resizer to resize image with ratio
• launched with python main.py --listen

Workflows

There are two scenarios: one object in a picture, and a stack of items in a picture. Therefore, I created two workflows.

• rmbg-multi-add-text

• rmbg-single-add-text

Problems

• I’ve introduced the “Image Crop” node to tailor the image containing only the object to its actual size. I haven’t yet found a better way to determine the object’s size automatically. I already created a github issue to ask if they can implement this feature. This is done by comfy-node-crop-image-by-mask