Batch to Image List
Split a batch into a list so ComfyUI processes frames one at a time
- images
- image
Batch to Image List is a two-line node that exists to fix a very specific ComfyUI problem: sometimes a big batch will eat your VRAM, and the escape hatch is to stop processing it as a batch. It takes an IMAGE tensor shaped [N, H, W, C] and splits it into a list of N individual frames. That's the entire job, and it's quietly one of the more useful utilities in the pack once you understand why it matters.
The mechanism is the point. ComfyUI treats an IMAGE as one tensor, so VAE-encoding or sampling a 64-frame batch can spike memory hard - the whole thing moves through the pipeline together. A list, by contrast, makes ComfyUI iterate: each frame travels through your graph on its own, so peak memory stays flat instead of scaling with the batch. The node's own docstring says it outright: splitting a batch avoids RAM spikes during VAE encode/decode. There's a matching ImageListToBatch node in the same pack to collect the results back into a single tensor when you're done.
When should you actually use it? The honest answer is: when something OOMs. If a batch of frames blows up at the VAE or at a per-frame filter, inserting this node before the heavy part and re-batching after is the classic fix. It's also the natural companion to the pack's list-oriented nodes - ImageListSampler for picking frames out of a list, LTXVMultiGuide for injecting reference images into an LTX latent - which all expect to work with lists.
Two things to know so it doesn't bite you. First, list output is contagious: every node downstream now runs per-frame, which is slower overall than batching (that's the price of lower memory). If a downstream node doesn't accept lists, ComfyUI will complain or quietly reconnect wrong - re-batch with ImageListToBatch before it. Second, the input is a single IMAGE input called images, and the single output is image - don't overthink the naming.
It's about as low-risk as custom nodes get: no parameters, no dependencies beyond torch, no model files. Install is just the pack - ComfyUI Manager, search "AnotherUtils", or cd ComfyUI/custom_nodes && git clone https://github.com/marcoc2/ComfyUI-AnotherUtils and restart. If you only ever use one node from this pack, this is a reasonable candidate, because it's a solution to a problem everyone with a big batch eventually meets.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |