Get Last Image

Get Last Image is the mirror-image of its sibling Get First Image, and it exists for the exact same reason: video batches are annoying to poke at. Give it an image batch [B,H,W,C] or a list of image tensors and it hands you back just the final frame as a clean single image, plus a count of how many frames went in.

When is the last frame the one you want? More often than you'd think. In a video generation chain, the last frame is the endpoint - the frame you'd use to seed the next clip if you're looping, the frame you'd feed back into an image-to-video pass to extend the sequence, or the frame a Smart LM image task should look at when you want "what does this chain end on?" rather than "summarize this video." The author's tooltip says it plainly: feed the final frame of a video/chain into an image task to avoid video-mode trimming. Same job as Get First, opposite end.

How it works

Nothing clever under the hood, and that's the point. Input gets flattened into frames, the node returns frames[-1], and the count output reports the total frames so you can verify you actually got a batch and not a single image. A lone image passes straight through. If the input is empty the node warns in the logs and returns a 1×64×64 black placeholder with count = 0 instead of throwing - convenient, but treat it as a red flag if you see black appear downstream.

The three sockets you'll touch

• image (in) - the batch or list.
• image (out) - one [1,H,W,C] frame: last in, nothing else.
• count (out) - total input frames. Wire it to a text/Show node when you're debugging.

The count output is genuinely useful for a workflow hygiene habit: it's the cheapest possible check that an upstream loader gave you the frame count you think it did. Cheap wins like that are why a node this dumb still earns a spot in a utility pack.

Installing it

Part of the ComfyUI_Eclipse pack (r-vage, RvTools lineage). Through ComfyUI Manager: Custom Nodes → search ComfyUI_Eclipse. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI_Eclipse

Restart, then look under Eclipse → Image → Batch Operations. Dependencies are standard (torch, numpy, Pillow, opencv-python - all things ComfyUI usually has); run pip install -r requirements.txt if a node errors on import. Keep in mind this pack targets current ComfyUI - the frontend-heavy parts want 1.51.2+, and its v4.0 rewrite dropped all legacy RvTools_v2 nodes, so ancient workflows may need the built-in migration tool.

Honestly, if you only ever need one of the pair, Get Last is the one I'd grab first - "where does this end" is the question video work keeps asking. Just remember both endpoints return the placeholder black image on empty input, and check your count.