Convert to List

Lots of ComfyUI nodes are built to chew through a whole batch of frames at once: one image tensor with N frames stacked in the batch dimension. But plenty of useful operations only make sense one frame at a time - a per-frame crop, a mask pass that should see a single image, a per-image comparison. Convert to List is the tool that opens the batch up, handing you a Python list where each item is one frame you can route independently.

What it does

The To List [Eclipse] node takes any batch - images, masks, latents, conditioning - and outputs a list of the individual items inside it. The clever bit is that it detects the type automatically via a dynamic match type, so there's no dropdown where you declare "this is an IMAGE batch" before it will work. Connect an image batch and you get a list of images. Connect a mask batch and you get a list of masks. It just figures it out from the wire.

This is the exact inverse of Convert to Batch in the same pack, and they're designed to be used together. The pattern that justifies both: take a batch, split it with Convert to List, process each element through nodes that only handle single items (and that conveniently emit lists themselves), then rejoin with Convert to Batch when the downstream node wants a tensor again. Open the batch, work on it, close it back up.

When you actually need it

The most common real-world trigger is video. Generate 81 frames as one batch, and suddenly you want to inspect frame 40, or run a stabilizer that wants individual images, or apply a per-frame color grade. Convert to List is also the escape hatch when a node is silently treating your whole batch as one image - some poorly-written nodes don't respect the batch dimension, and splitting first is the workaround that doesn't require fixing the node.

Inputs and outputs

One input, input: any value or batch to convert. One output: the list format. As with its sibling, there are no settings to configure - it's genuinely fire-and-forget. The only real failure mode is feeding it a single item instead of a batch; if you connect one image expecting a list of one, check whether the upstream node is actually outputting a batch or just a lone tensor.

Install it

Part of ComfyUI_Eclipse, so the install is shared across all the [Eclipse] nodes. Easiest: ComfyUI Manager → Custom Nodes → search "Eclipse" → Install → restart ComfyUI. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI_Eclipse

Then restart. The pack's extra Python deps (opencv-python, pilgram, PyYAML, aiohttp) are installed by most setups; if you see import errors, run pip install -r custom_nodes/ComfyUI_Eclipse/requirements.txt in the same environment that runs ComfyUI. The node lives under Eclipse → Conversion.

Gotchas

Worth saying plainly: a list of frames is not a batch, and lots of nodes refuse to accept a list where they expect a tensor. If you split a batch and then wire the list straight into a node that wants an IMAGE, it'll complain - that's what Convert to Batch is for, on the other side. And if you're splitting a video batch to process frames individually, remember the order matters: the list preserves batch order, so as long as you rejoin in order you stay aligned with your audio and your frame indexing. Get that wrong and you'll get a video where the frames are subtly out of order and it'll take you an hour to notice.