⭐ Star Show Last Frame

This is the definition of a utility node: one input, one output, one idea. Star Show Last Frame takes an image batch - a decoded video, a sequence of frames, whatever - and hands you back just the last frame. That's the entire job, and honestly that's why it's useful. It's the node you grab when you need "the ending" of a sequence and don't want to write a slice node or count frames by hand.

When you actually reach for it

The most common case is last-frame conditioning for video models. LTX and a few other pipelines want a final frame as an anchor for extension or loop-closure work, and being able to peel the last frame off a decoded batch and route it straight into an image node is the cleanest way to do that. It also earns its keep as a sanity check: after you decode a video to frames, drop this on the tail and you instantly see what the video actually ends on - great for catching a last-frame glitch before you save or loop it.

The sibling you should know about: Star LTXV Get Last Frame does this for latents inside an LTX pipeline. This node works on decoded IMAGE tensors, so use it after a VAE decode. They solve the same problem on different sides of the video.

The interface

• images (IMAGE) - the batch.
• image (IMAGE, output) - the final frame, as a single-image batch.

Mechanically it's a slice: images[-1:] with a clone, so the output is always a valid IMAGE tensor with one frame. An empty or single-frame input just passes through as-is, which is the sort of defensive behavior you want from a tiny node you'll wire into pipelines you don't control.

Installing

It comes with the StarNodes pack, same as everything else:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

Restart ComfyUI, then find it by searching star - it lives under ⭐StarNodes/Helpers And Tools. Manager users can install by searching "Starnodes".

Notes

No heavy dependencies, no models to download, nothing to tune. The only thing that trips people up is expecting it to work on latents - it won't, that's the LTXV node's job. Feed it an IMAGE batch and it's boring and reliable, which for a helper node is the highest compliment it can get.