Select Indexes From Images

You loaded a 300-frame clip and you need frames 10–20, or every 4th frame, or just the last one. Most ComfyUI frame handling makes you either take everything or write tedious per-index nodes. Select Indexes From Images gives you a single text field with one compact syntax that does all of it. It's the frame-filtering workhorse of SineSwiper's LoadAnim-Adv pack, and once you've used it, the stock way feels broken.

The one input that matters: the syntax

Everything about this node is the indexes_to_select string. It's Python's range() written with colons - first frame is index 0, negative numbers count from the end:

• 5 - frame 5 only
• 0:5 - frames 0 through 4 (end exclusive)
• 1: - frame 1 to the end
• :-5 - everything except the last 5 frames
• ::2 - every second frame
• 0:10:2 - frames 0, 2, 4, 6, 8
• -1 - the last frame
• 0,5:10,-1 - mix singles, ranges, and the tail, separated by commas

Empty :: on its own is a full slice, and the default value is 0: - which is "everything from frame 0 on," i.e., pass-through. If you're coming from Python's range, this is that, with negatives working the way you'd hope.

How it works

The node takes an IMAGE tensor - which in ComfyUI is laid out as [frames, height, width, channels] - and indexes along the frame axis. Feed it a still image and it's a one-element tensor, so any valid selection is a no-op unless you pick 0. Feed it a multi-frame tensor from a loader in this pack (or any animated source) and you get back a new tensor holding only the selected frames, same order as written. Empty or invalid selections raise an error rather than silently returning nothing.

When to reach for it

• After a directory loader, to trim each file down to the frames that matter.
• Before a video-model pipeline that wants a specific frame count or a stride - ::2 is an easy way to halve a clip for faster sampling.
• For extracting a single keyframe (-1 for the last frame of a render) to feed a refiner or an upscaler.

Inputs and outputs

Just two inputs: images (an IMAGE tensor) and indexes_to_select (the string). One output: images, the selected set. That's the whole node - refreshingly small, which is exactly right for a node whose only job is slicing.

Installing

Part of ComfyUI-LoadAnim-Adv by SineSwiper. ComfyUI Manager → search "LoadAnim", or:

cd ComfyUI/custom_nodes
git clone https://github.com/SineSwiper/ComfyUI-LoadAnim-Adv.git

Restart ComfyUI. No models, no extra dependencies beyond what a stock install already has (torch, numpy, Pillow).

Gotchas

• End-exclusive ranges catch everyone. 0:10 gives you 10 frames (0–9), not 11. If you want frames 1 through 10 inclusive, write 1:11.
• Negative selection is "from the end," not "from the start going down" - -1 is the last frame, and you can't select a frame "at index -2 in the front."
• This node expects a single tensor. If your upstream sends a list of tensors (the directory loader with flatten_frames off), it handles that too - it runs the selection on each set - but for a flat single tensor, this is all you need.

One text field, one syntax, and suddenly frame cherry-picking isn't a chore. Learn the syntax once and it carries across the whole pack - the directory loader and Select Indexes From Any speak the same language.