ComfyUI LatLong — Equirectangular Image Nodes

High‑quality equirectangular (360°) image processing nodes for ComfyUI. Rotate with yaw/pitch/roll, adjust the horizon, crop to 180° or square, convert to cubemaps, and extract perspective views — all designed for panoramic workflows.

What’s New

• All‑in‑One node now embeds controls directly in the node (sliders/toggles/dropdowns). No external parameter nodes are required for rotation/crop/interpolation.
• Helpful tooltips added to every node and parameter to clarify behavior and best‑use guidance.

Installation

Option A — Clone (recommended)

• From ComfyUI/custom_nodes:
  • git clone https://github.com/cedarconnor/comfyui-LatLong.git
  • cd comfyui-LatLong
  • pip install -r requirements.txt
  • Restart ComfyUI

Option B — Manual

• Extract the repo into ComfyUI/custom_nodes/comfyui-LatLong/
• Install deps: pip install numpy>=1.21.0 opencv-python>=4.5.0 scipy>=1.7.0 torch>=1.9.0
• Restart ComfyUI

Requirements

• ComfyUI (latest), Python 3.8+
• Python packages:
  • numpy>=1.21.0, opencv-python>=4.5.0, scipy>=1.7.0, torch>=1.9.0

GPU Notes

• Some GPU paths use torch.grid_sample and support bilinear/nearest only. Lanczos/Bicubic use CPU (OpenCV) paths.

Nodes Overview

1. Equirectangular Rotate

• Purpose: Rotate an equirectangular image with yaw/pitch/roll and adjust the horizon.
• Inputs:
  • image: equirectangular tensor (B,H,W,C) in [0,1]
  • yaw_rotation: horizontal rotation (degrees)
  • pitch_rotation: vertical tilt (degrees)
  • roll_rotation: bank rotation (degrees)
  • horizon_offset: vertical horizon shift (degrees)
  • interpolation: lanczos | bicubic | bilinear | nearest
  • backend: auto | cpu | gpu

2. Equirectangular Crop 180

• Purpose: Extract a horizontal FOV (default 180°) window from an equirectangular image.
• Inputs:
  • image
  • output_width, output_height: target size
  • maintain_aspect: preserve natural aspect ratio for selected FOV
  • center_longitude_deg: FOV center longitude (degrees)
  • fov_degrees: horizontal field of view (degrees)
  • interpolation

3. Equirectangular Crop Square

• Purpose: Center‑crop width to match the original height (perfect square).
• Inputs: image, interpolation (not used by pure crop; kept for consistency)

4. Equirectangular Processor (All‑in‑One)

• Purpose: Rotate then optionally crop to 180° or to a centered square — a single‑node workflow. Note: square crop takes precedence over 180°.
• Inputs:
  • image
  • yaw_rotation, pitch_rotation, roll_rotation, horizon_offset
  • crop_to_180, crop_to_square
  • output_width, output_height (for 180° crop)
  • interpolation

5. Equirectangular Perspective Extract

• Purpose: Extract a pinhole‑camera perspective view (yaw/pitch/roll/FOV) from an equirectangular panorama.
• Inputs: image, yaw_rotation, pitch_rotation, roll_rotation, fov_degrees, output_width, output_height, interpolation, backend

6. Equirectangular → Cubemap (3×2)

• Purpose: Convert equirectangular panorama to a 3×2 cubemap atlas.
• Inputs: image, face_size, layout (3×2), interpolation
• Face order: [left, front, right] on top row; [back, top, bottom] on bottom row

Quickstart Examples

Rotate

• Add Equirectangular Rotate, connect an image, tweak yaw/pitch/roll/horizon.

Square Crop

• Add Equirectangular Crop Square to center‑crop to a perfect square.

180° Extraction

• Add Equirectangular Crop 180, set center_longitude_deg and fov_degrees as needed, enable Maintain Aspect for correct proportions.

Perspective View

• Add Equirectangular Perspective Extract, set yaw/pitch/roll, FOV, and output size.

Cubemap

• Add Equirectangular → Cubemap (3×2), choose face_size, select interpolation.

All‑in‑One

• Add Equirectangular Processor (All‑in‑One), adjust rotation; enable square or 180° crop; pick interpolation. All controls are embedded in the node.

Technical Notes

Coordinate System

• Accurate spherical transforms with longitude wrapping and pole handling.

Interpolation

• CPU: OpenCV remap (Lanczos, Bicubic, Bilinear, Nearest)
• GPU: torch.grid_sample (Bilinear/Nearest where applicable)

Pipeline

• Input (B,H,W,C) → float32 processing → transform → resample → output tensor in [0,1].

Troubleshooting

• Blurry output: prefer Lanczos/Bicubic for final renders.
• Memory limits: reduce resolution or batch size.
• GPU path missing filters: use CPU for Lanczos/Bicubic.

Contributing

Issues, feature requests, and PRs are welcome.

License

Open source — see the repository license.

— Ideal for 360° photography, VR, panoramic pipelines, and social content where precise equirectangular control matters.