Load Image (RMBG)

The Load Image (RMBG) node (AILab_LoadImage) loads an image from disk (or a configured directory) into ComfyUI and optionally extracts a mask from one of its color channels, providing an efficient entry point for background‑removal and segmentation workflows.

Overview

Load Image (RMBG) is part of the ComfyUI‑RMBG suite and is tailored for scenarios where an input image and a channel‑derived mask are both needed, such as subject isolation, matting, or pre‑processing for RMBG‑2.0 background‑removal models. It loads the selected image file, can rescale it according to several modes, and extracts a binary or grayscale mask from a chosen channel (alpha, red, green, or blue), also producing an RGB visualization of that mask plus the image dimensions. These outputs connect directly to RMBG, segmentation, compositing, and inpainting nodes in downstream pipelines.

Visual Example

Official Documentation Link

https://comfyai.run/documentation/AILab_LoadImage
https://github.com/1038lab/ComfyUI-RMBG

Inputs

Outputs

Usage Instructions

Select Load Image (RMBG) from the AILab/RMBG category and choose an image filename from the dropdown list of available files in your configured directory. Pick the mask_channel to indicate where the subject mask resides (for example alpha for images with transparency or a color channel for chroma‑key–style masks), then adjust scale_by, resize_mode, and size if you want to normalize resolution for downstream models. Connect IMAGE and MASK to RMBG, segmentation, or compositing nodes, and optionally use MASK_IMAGE plus WIDTH/HEIGHT for previews, overlays, or assertions about input size.

Advanced Usage

In advanced background‑removal workflows, you can feed MASK from Load Image (RMBG) directly into RMBG or LayerStyle nodes to combine channel‑based masks (for example provided by a 3D renderer) with AI‑predicted masks for more controlled composites. When preparing training or evaluation sets, the node can standardize resolution via resize_mode and size, ensuring every image/mask pair conforms to the input requirements of downstream networks. For pipelines that mix multiple mask strategies (alpha, green‑screen, segmentation), chaining multiple Load Image (RMBG) instances or combining their masks with math/mask‑merge nodes enables complex region selection before final matting or relighting.

Example JSON for API or Workflow Export

{
   "id":"ailab_load_image_1",
   "type":"AILab_LoadImage",
   "inputs":{
      "image":"sample_input.png",
      "mask_channel":"alpha",
      "scale_by":1.0,
      "resize_mode":"longest_side",
      "size":0
   }
}

Tips

• Use mask_channel = alpha when working with PNGs or formats that already contain a clean transparency mask; this often yields better results than re‑segmenting from RGB.
• For green‑screen or chroma‑key footage, experiment with red, green, or blue channels to derive a simple thresholdable mask before refining with AI‑based RMBG nodes.
• Keep scale_by near 1.0 and rely on resize_mode + size to adjust resolution uniformly across a dataset, which helps preserve aspect ratio and consistent framing.
• Use MASK_IMAGE in a side‑by‑side viewer or stitched reference panel to quickly inspect mask quality before running heavier background‑removal or compositing pipelines.
• Organize your image directory clearly (for example by project or input type) so the image dropdown remains easy to navigate in large production setups.

How It Works (Technical)

Load Image (RMBG) reads the selected file from disk, converts it into an internal tensor representation, and applies optional resizing based on scale_by, resize_mode, and size, typically preserving aspect ratio as configured. It then selects the requested mask_channel: for alpha, it extracts the alpha plane; for red, green, or blue, it slices the corresponding channel from the RGB data and normalizes it into a mask tensor. A separate MASK_IMAGE is generated by mapping the mask into an RGB visualization (for example grayscale), while WIDTH and HEIGHT are computed from the final image dimensions. All tensors are returned in ComfyUI’s IMAGE and MASK formats so they can be consumed directly by RMBG and segmentation nodes.

Github alternatives

FAQ

1. Do I need RMBG models installed to use Load Image (RMBG)?
No, Load Image (RMBG) itself only loads images and extracts masks from channels; however, it is typically used together with RMBG or segmentation models that consume those outputs.

2. What is the difference between MASK and MASK_IMAGE?
MASK is a single‑channel tensor intended for processing (matting, compositing, model inputs), while MASK_IMAGE is an RGB visualization of the mask that is easier to preview and overlay in image‑space workflows.

3. How should I choose resize settings for best results?
Use resize_mode = longest_side or shortest_side with a size that matches your downstream model’s expected resolution, and keep scale_by at 1.0 unless you need an extra global scale adjustment.

Common Mistakes and Troubleshooting

A common pitfall is selecting a mask_channel that does not actually contain meaningful mask data (for example alpha on a JPEG or a color channel with low contrast), resulting in nearly empty or noisy masks; in such cases, inspect MASK_IMAGE and adjust the channel choice or your source images. Another frequent issue is mismatched resizing relative to downstream models: overly aggressive scaling or inappropriate resize_mode can distort subjects or reduce segmentation accuracy, so align your size settings with the requirements of RMBG or segmentation networks. If the node fails to list images in the dropdown, verify that the configured directory exists, that the file formats are supported (PNG, JPG, etc.), and that paths match your ComfyUI‑RMBG installation; missing or mis‑installed RMBG packages can also prevent the node from registering correctly.

Conclusion

Load Image (RMBG) is a powerful entry node for background‑aware workflows in ComfyUI, combining robust image loading, flexible resizing, and channel‑based mask extraction to streamline background removal, segmentation, and advanced compositing pipelines for both creative and production use.