ImageToBatch

ComfyUI is quietly full of nodes that take a whole batch and a few that let you put one thing into a batch. ImageToBatch is one of the few: it replaces or inserts an image (or a small batch) into an existing image batch at a chosen index. It's a plumbing node - nobody builds a workflow around it, but when you need it, you really need it.

The classic scenario: you've got a batch of generated images heading into a ControlNet, an upscaler, or a comparison grid, and you want to swap one frame, inject a reference image at slot zero, or append a new sample. Without this, your options are hacky concat-then-slice games. With it, you set an index and pick a mode: replace overwrites an existing slot (batch size stays the same), insert shifts everything to the right and grows the batch. The replace-by-default change landed in version 1.2.52 - before that it always inserted, which caught people off guard, so check which mode you're in.

How it works

Under the hood it's a tensor splice along the batch dimension. The incoming image can itself be a batch of several images, and the node places all of them starting at index. The constraints are the honest ones: spatial shape, channel count, dtype, and device must all match the target batch, because there's no resizing or conversion happening here - it's placement, not adaptation.

The inputs that matter

• image_batch - the batch you're modifying.
• image - what you're placing (single image or small batch).
• index - where it goes. In replace mode it must point at an existing item; in insert mode it can equal the current batch size to append.
• mode - replace (overwrite, same size) or insert (shift right, grows).

Output is the modified IMAGE batch.

Installing it

Part of Skoogeer-Noise. Manager → search "Skoogeer-Noise", or:

cd ComfyUI/custom_nodes
git clone https://github.com/ttulttul/Skoogeer-Noise

Restart ComfyUI. Deps are torch, numpy>=1.26, einops, pyyaml>=6.0.3; no model downloads.

Common gotchas

The error you'll actually hit is a shape mismatch: drop a 1024×1024 image into a batch of 512×512s and it fails loudly, because the node won't resize for you. Resize upstream. Also mind the index semantics between modes - replace at an index past the end errors, while insert at batch_size is a valid append. And if a workflow you downloaded predates 1.2.52, its ImageToBatch may rely on the old always-insert behavior; re-check the mode dropdown if the batch size isn't what you expect.