Combine images

This one does exactly what the display name says and nothing more: two IMAGE batches go in, one bigger batch comes out. The name doesn't scream "multi-GPU distributed rendering pack," which is probably why it pulls more generic search traffic than the rest of ComfyUI_NetDist despite living in a pretty niche corner of the ecosystem - most people who find it aren't running two GPUs at all, they just want to merge two batches.

Inside NetDist itself, it's the node you drop right after FetchRemote: your local batch plus the batch that came back from the other instance, stitched into one so a single Save Image or Preview Image at the end shows everything together instead of two separate result strips.

How it works

There's no tooltip on this one and nothing exotic exposed in its schema - it's a straight concatenation along the batch dimension. Same resolution, same channel count in, one combined batch out. If images_a and images_b don't match in shape (different resolutions, most commonly), expect a tensor-shape error rather than silent resizing - this node isn't doing any resampling for you.

The inputs and outputs that matter

• images_a, images_b (IMAGE, both required) - the two batches to merge. Order doesn't functionally matter; it's a concat, not a blend.
• images (IMAGE, output) - the combined batch, same count as images_a plus images_b together.

That's the entire interface. No batch-size limits beyond whatever your downstream node can handle, no options to control ordering or interleaving.

How to install it

Via ComfyUI Manager: search ComfyUI_NetDist. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/city96/ComfyUI_NetDist
pip install requests

Restart ComfyUI. The whole pack's only dependency is requests, so grabbing this one small node doesn't cost you anything heavy even if you never touch the remote-execution side.

Common issues & troubleshooting

Shape mismatch errors. The two batches need matching resolution (and ideally the same aspect ratio going in) - if your local render came out at 1024x1024 and the remote batch was 768x768, this will error rather than pad or crop. Resize one side to match before combining.

You installed a whole distributed-rendering pack for one combine node. That's a fair reaction if you found this purely through search and have no interest in running a second GPU - but the pack's footprint is genuinely light (one Python dependency, no models), so there's no real cost to having it installed just for this.

Using it outside NetDist entirely. Nothing about CombineImageBatch references anything NetDist-specific - it works fine for merging any two IMAGE batches from two separate generation paths in the same graph, like a base batch and an upscaled-variant batch you want previewed side by side.