❌ 删除首/末帧图片

The mirror image of GetFirstImageFromBatch: where that node extracts a frame, this one removes it. "❌ 删除首/末帧图片" takes an IMAGE batch, drops the first or last frame, and returns the rest. It sounds like a nothing-node until you want to trim a duplicated frame, drop the fade-in/out of a generated video, or - in the batch-assembly pattern this pack loves - progressively consume a stack one frame at a time.

How it works

A slice, nothing more. In first mode it returns image_batch[1:]; in last mode, image_batch[:-1]. Two edge cases get handled deliberately: if the batch is already empty, it passes through unchanged, and if you feed it a single 3D image, it returns a zero-size batch rather than crashing. The latter is worth knowing - remove the only frame and you get empty, not an error, which can quietly surprise a downstream node that assumed it'd get something.

Inputs and output

• image_batch (IMAGE) - the stack to trim.
• mode - first or last, default first.
• Output: image_batch (IMAGE) - the batch minus one frame, batch dimension preserved.

The pairing that makes it useful

On its own it trims frames. In combination it becomes a loop primitive: GetFirstImageFromBatch to grab the current first frame, this node to drop it, append the result back with AppendImagesToBatch - and you've got a "deal one card at a time" structure you can drive around a loop or a reroute. That's the pattern these batch utilities in the pack are clearly designed to support, even if the README never says so.

Real-world trims it's also good for: your video generator always emits the same title card on frame one and you want it gone; you're building a "last frame" workflow and need to drop the final frame to make room for a different ending.

Installing it

Standard pack install - ComfyUI Manager → search comfyUI_LLM, or:

cd ComfyUI/custom_nodes
git clone https://github.com/XieJunchen/comfyUI_LLM
# restart ComfyUI

Pure torch slicing, no extra dependencies.

Where people get burned

• Single-frame input → empty output, silently. No error, just a (0,...) tensor that many nodes can't handle. If your downstream suddenly produces nothing, check whether the batch shrank to zero.
• It does not check for empty input. A zero-length batch in gives a zero-length batch out - no crash, but also no work done. Harmless, but easy to misread as a bug.
• No resolution checking, no mode beyond first/last. Nothing to tune, which keeps it honest.

It's a one-trick node with one good trick. If you're assembling or trimming frame batches in this pack's ecosystem, it earns its slot next to the other batch utilities.