ComfyUI OA 360 Clip

A ComfyUI custom node for extracting perspective views from 360° equirectangular images with an interactive drag-and-drop crop interface.

Why? Because I want to do camera shots from different angles with a consistent background. One way to do that is to generate a 360 image, then take background snips from that image. But you need to undo the curviture introduced by 360 photos, hence this extension.

Why the OA prefix? I go by "Ordinary Alan" (username: xordinary99) on the ComfyUI Discord server.

Features

• 360° Image Support: Extract perspective views from equirectangular 360° images using gnomonic projection
• Interactive Crop Zone: Drag and resize the crop region directly in the node preview using the projected outline
• Visual Feedback: See the projection outline on the input image to understand what will be extracted
• Flexible Output Dimensions: Set output width and height either via external inputs or through the UI widgets
• Resizable Node: The node can be resized to fit your workflow, and the preview scales to fit the available space

Installation

Place this folder in your custom_nodes directory:

ComfyUI/custom_nodes/ComfyUI-OA-360-Clip/

Restart ComfyUI to load the extension.

Usage

1. Add the OA 360 Clip node to your workflow
2. Connect a 360° equirectangular image to the image input
3. Run the workflow once to load the image preview
4. Adjust the crop region by:
   • Dragging the crop outline: Click and drag inside the projected outline to move the crop region
   • Resizing: Click and drag the corner handles to resize the crop region (maintains output aspect ratio)
   • Jumping: Click outside the crop outline to instantly move the crop center to that location
5. Set the output dimensions:
   • Use the output_width and output_height widgets in the node
   • Or connect external integer inputs to these properties
6. Run the workflow to generate the output image

Node Properties

Inputs

• image: Input 360° equirectangular image (required)
• output_width: Width of output image (optional, default: 1280, range: 1-8192)
• output_height: Height of output image (optional, default: 720, range: 1-8192)

Outputs

• IMAGE: The extracted perspective view as a standard ComfyUI image tensor

Crop Region

The crop region is controlled entirely through the interactive UI:

• Center-based: The crop is defined by its center point (crop_center_x, crop_center_y) and dimensions (crop_width, crop_height)
• Maximum width: The crop width is limited to 25% of the input image width to prevent excessive distortion
• Wrapping: The center point can wrap around horizontally (equirectangular images wrap at the edges)
• Aspect ratio: When resizing by dragging corners, the crop maintains the aspect ratio of the output dimensions

Notes

• The crop region defines the area in the equirectangular image to extract
• The output dimensions define the final perspective view size
• The projection uses gnomonic projection, which may have distortion near the poles
• For best results, keep crop regions away from the extreme poles (±90° latitude)
• The node can be resized by dragging the bottom-right corner
• Crop values are automatically saved with the workflow

Technical Details

• Projection: Gnomonic (pinhole) projection for perspective view extraction
• Coordinate System: Equirectangular input with spherical coordinate conversion
• Crop Constraints: Maximum crop width is 25% of input image width
• State Management: Crop values are stored in node properties and synchronized with the backend via API

Acknowledgements

This project was inspired by the ComfyUI-Olm-DragCrop extension, which provides an excellent interactive crop interface for ComfyUI. However, this is a clean implementation written from scratch, as the original extension's license was not permissive enough for our needs. This project was vibe coded with Cursor AI from scratch (and it took a while even then to get right!)

If you need to clip a rectangular region from a standard (non-360°) image, we recommend using the ComfyUI-Olm-DragCrop extension instead, which is specifically designed for that use case.