StableZero123_Conditioning_Batched_motorway_edition

This is StableZero123_Conditioning with a batch superpower: instead of producing one view, it produces a whole batch, and it can increment the camera angle between batch items. So you set it to 8 frames, give it a starting elevation and azimuth, and each item in the batch gets a slightly rotated view of your object. The Motorway edition pulls the three inputs from the pipe by key and stores the outputs back under keys.

If you've seen Stable Zero123 results that look like a turntable of an object, this node is how that was built. The single-view version is great for one angle; the batched version is what you use when you want the orbit - feed the batch of latents to the sampler and you get eight views of the same subject at 45-degree-ish steps, which you can then assemble into a spin or just pick the best frame from.

What to set

• INPUT_clip_vision_key / INPUT_init_image_key / INPUT_vae_key (defaults clip_vision / init_image / vae) - the same three inputs as the non-batched node.
• width / height (default 256 × 256), batch_size (default 1) - size and frame count.
• elevation / azimuth (defaults 0) - the starting camera angle.
• elevation_batch_increment / azimuth_batch_increment (defaults 0) - how many degrees to shift the camera per batch item. To orbit horizontally, set azimuth_batch_increment to something like 45 and batch_size to 8.
• OUTPUT_positive_key / OUTPUT_negative_key (defaults positive / negative) - conditioning output keys.
• OUTPUT_latent_key (default latent) - the batched latent output key.

Output is the MOTORWAY with positive, negative, and a batched latent stored.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart; it's under agilly1989 Nodes → Motorway-ed → conditioning → 3d_models. Manager installs the pack too. No pip dependencies.

Where people get burned

Increments of 0 do nothing - set azimuth_batch_increment to 0 and you're just rendering the same angle 8 times, which burns VRAM for nothing. And watch total VRAM: batch_size × sampler cost adds up fast on a 256px latent. Render at 256 and upscale after, don't inflate the batch latent size.

Same model-pairing rule as the single-view version - CLIP vision, VAE, and checkpoint must all be the Zero123 family. And the pack standard: keys are case-sensitive and exact; 'MotorwayClass' object has no attribute 'hash_' means a key isn't in the pipe. Beta pack, motorway-ed nodes come and go across releases - a disappearing node after an update is upstream churn, not you.