🔍 获取首/末帧图片

Almost every video or animation workflow eventually asks the same question: what does the first frame look like? The first frame is your thumbnail, your img2img seed, your reference card for an LLM describing the scene. This node, "🔍 获取首/末帧图片", answers it directly - it slices the first (or last) frame out of an IMAGE batch and hands it back as a single-image batch, with the width and height thrown in as a bonus.

How it works

Trivial in the best way: it takes the batch and returns image_batch[0:1] (or image_batch[-1:] if mode is last). The slice keeps the batch dimension - you get a (1, H, W, C) tensor, which is exactly what every node downstream expects, since most ComfyUI nodes refuse plain 3D images. A 3D single image input gets promoted to a batch of one instead of erroring. It also unpacks width and height from the frame's shape while it's in there.

Inputs and outputs

• image_batch (IMAGE) - the stack to slice from.
• mode - first or last, default first.
• Outputs: image (IMAGE) - the single extracted frame; width (INT) and height (INT) - the frame's dimensions.

Where it fits

The natural wiring in this pack: after SplitVideoByFrames hands you clips, or after any video loader gives you a frame batch, grab the first frame and route it to an img2img node, an image-to-video model, or a captioning LLM. The width/height outputs are the quiet win - instead of dragging a separate Get Image Size node into the graph, you get dimensions for free, which is handy for wiring up size-conditional logic.

One pairing that shows up a lot in batch-assembly loops: GetFirstImageFromBatch to pull the current first frame, and its sibling RemoveFirstOrLastImageFromBatch to drop it - together they let you peel a batch like a deck of cards.

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.

The edge cases

• Empty batch → hard error. It raises "image_batch is empty!" rather than returning garbage. Fair, but know it's coming if you feed it the output of an empty CreateEmptyImageBatch.
• No middle mode. Just first and last - if you need the Nth frame, GetVideoClipByIndex or an image-indexing node from the wider ecosystem is your tool.

It's a two-option utility node, and it's fine that it's boring - half the battle in ComfyUI is wrangling batch shapes, and this node removes the fiddliest slice for you.