Street View Equirectangular Loader

If you want a full sphere - the whole scene around a point, sky included - the StreetViewCubemapLoader gives you six flat faces and this node gives you the thing you actually hand to a VR headset or a 360 viewer. StreetViewEquirectangularLoader fetches six Street View captures and projects them into a single 360°×180° equirectangular image: the familiar 2:1 panorama where every direction is stretched flat into one rectangular picture.

Think virtual tours, real-estate walkthroughs, 360 video, or environment lighting for a render where you care about the sky. It's the same family as the cubemap node, but where that one hands you raw faces to assemble, this one does the projection for you.

How it works

Six API calls - front, right, back, left, top, and bottom - each at 90° field of view. Street View's camera orientation is quirky, so the node applies per-face flips and rotations before stitching anything together, then uses vectorized numpy math to map every pixel of each cube face onto the sphere and flatten it into the 2:1 output. The source keeps the whole thing reasonably fast, but this is still the slowest node in the pack - the projection is real math, not a paste job.

The inputs that matter

• location - "lat,lon", same as the rest of the pack.
• face_resolution - 256, 512, or 640 per face. 640 is the API ceiling.
• upscale_factor (1–4) - this is the quality knob. The node upscales each face before projecting it, which cuts the quality loss that the geometric transformation would otherwise introduce. The default of 2 doubles each face (640 → 1280) and is a solid starting point.
• upscale_method - how that pre-upscale is done: LANCZOS (default, best quality) through NEAREST (fastest, soft). Lanczos is the one to leave it on.
• interpolation_mode - how pixels are sampled during the projection itself: BILINEAR (smoother, default) or NEAREST (sharper but prone to aliasing).
• historical_date_id (optional) - one honest caveat here: this is the single input in the whole pack with an extra prerequisite. The tooltip says a historical equirect requires the Street View Image Metadata API to also be enabled on your Google Cloud project. Current-date panoramas need nothing extra.

Outputs: equirectangular_image (the main 2:1 IMAGE), the six individual post-transform faces (front, back, left, right, top, bottom), and metadata (STRING) reporting how many of the 6 faces fetched successfully plus the request URLs. If a face fails, it becomes a gray placeholder - same behavior as the cubemap node, same advice: read the metadata.

Two honest costs

Six API calls per run, like the cubemap node. Plus, at high face resolutions and upscale factors, this one is both slower and heavier on RAM than anything else in the pack - the upscaled source faces plus the projection working set add up. Start at 512 or 640 with a 2× factor, check the quality, then decide if you need more.

Install

Standard for this pack: clone ru4ls/ComfyUI_StreetView-Loader into custom_nodes, pip install -r requirements.txt, restart ComfyUI, and set your Street View Static API key in the pack's .env. Node lives under Ru4ls/StreetView. No models to download; the only real requirement is the Google key and a little patience while it projects.