Face to Equirectangular

Face to Equirectangular (Face2E) is the return trip from the pack's single-face workflow. You pulled one face out of a panorama, edited it - inpainted the tripod hole out of the Down face, redrew the sky on Up, upscaled Front - and now you need it back in equirectangular form. This node takes a single flat face image and rebuilds a full panorama around it, where every other face is a flat constant color.

Notice what that means: the output is not your original panorama with the edited face dropped in. It's a new pano where one face is your image and the other five are a flat fill (default black). The README's framing is honest about it - this is a building block. The intended pattern is to generate a face in isolation and then either use it as-is or composite it back over your original.

Inputs

• image - your flat face image.
• face - which face it represents: Up, Down, Front, Right, Left, or Back (default Down, which tells you the author's intended use case: fixing the nadir).
• base_equi_color - the flat fill color for the other five faces, default 0.0 (black). If you want a neutral background for generation, set a gray value here.
• padding_mode - sampling mode for the conversion, bilinear default.

Output is Equirectangular Image - the rebuilt pano.

How it works

It builds a cubemap dict with your image on the chosen face and base_equi_color on the rest, then runs c2e from pytorch360convert to project the whole cube back to a 2:1 equirect. Same math as the pack's full Cubemap to Equirectangular node, just with five faces that are flat color.

The workflow in practice

The classic move: extract the Down face with Equirectangular to Face, inpaint/regenerate it (or generate a whole new ground plane), then Face2E it back with a base_equi_color that matches your pano's general ground tone. You'll get a full pano that's your new ground plus a flat-colored sky - and then you composite that over the original (masked or blended) so only the ground region actually changes. If you want to keep things surgical, the mask twins (Equirectangular Mask to Face / Face Mask to Equirectangular) do the same trip for MASK tensors, so the inpaint mask stays aligned through the round trip.

The flat background is the gotcha to keep in mind: if you just swap the Face2E output straight in as your final image, the other five faces being flat black is going to look broken. Treat it as an intermediate - a canvas or a composite input - not as a finished panorama.

Install

Pack install as usual: ComfyUI Manager → "ComfyUI_pytorch360convert", or git clone https://github.com/ProGamerGov/ComfyUI_pytorch360convert into ComfyUI/custom_nodes and restart. The requirements.txt is empty, so python -m pip install pytorch360convert is likely needed by hand. No model downloads.