⭐ Star LTXV Get Last Frame

A decoded video in ComfyUI is just a batch of frames, and half the useful tricks in video work come from pulling one frame out of that batch and doing something with it. Star LTXV Get Last Frame is the smallest possible version of that trick: it takes an IMAGE batch and outputs the final frame as its own IMAGE. One input, one output, nothing else - and it earns its place in any LTX-Video workflow that needs a loop or a continuation.

Why you'd reach for it

The headline use is seamless looping video. Generate a clip with LTX, decode it, grab the last frame, and feed that frame back as the first frame (or a conditioning reference) of the next generation - the loop now starts where it ended, so two clips weld together instead of visibly jumping. That's a classic LTX workflow pattern, and this node is the "close the loop" step in it.

The second use is continuation: when you want the story to keep going past the end of a clip, the last frame is exactly what you should reference. LTX-Video's whole family is built around conditioning on images for image-to-video, so having the last frame extracted cleanly means you can chain clips without a manual "save frame, load frame" detour.

The source even defends against the common mistake of feeding it the wrong thing: if you pass a dict where an IMAGE tensor is expected, it raises a clear error naming the keys it found, instead of crashing silently. That's the kind of defensive error message that saves you twenty minutes of head-scratching.

Inputs and outputs

• images (IMAGE) - the decoded frame batch, typically the output of a VAE decode or a video loader.
• last_frame (IMAGE) - just the final frame, as a single-frame batch.

Because the output is a standard IMAGE, it drops into Load-Image-style inputs everywhere: an image-to-video conditioning node, a VAE encode for the next pass, or a Preview node to eyeball it.

Installing

Part of the StarNodes pack - install Starnodes via ComfyUI Manager, or:

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

Restart, then search star on the canvas. No models or extra dependencies - it's a torch slicing operation.

Common issues

Not much to go wrong with a two-line slice, but two things trip people up. First, "last frame" is the last frame of whatever batch you feed it - if your video loader already trimmed or reordered frames, that's the frame you get. Second, if you don't want the very last frame (because it's a fade-to-black end card), grab an earlier one with a frame-picker node instead - this node has no offset control by design. And as ever with LTX, remember the decoded batch is at the model's internal half-resolution unless you upscaled after decode.