Wan Video Extract Frame (QQ)

Every once in a while you just want one frame out of a clip - the last frame of a generated video to chain into the next chunk, the first frame to use as a reference image, a mid-frame to inspect a motion problem. WanVideoExtractFrame is the smallest node in the siraxe ComfyUI-WanVideoWrapper_QQ pack (now ComfyUI-SA-Nodes-QQ) and it does precisely that: pull a single frame at an offset from either the start or the end of a video tensor and hand it back as a regular image.

It's a one-trick node, and that's fine. In a video workflow the thing you'll reach for it for most is ending a clip where the next one begins - extract the last frame of chunk one, feed it as the first-frame conditioning for chunk two. That "last frame of the previous chunk" pattern is everywhere in long-form Wan work, and this node's first_frame toggle exists so you don't have to count frames from the front when you want the tail.

How it works

The node accepts a video as an IMAGE tensor in either (T, H, W, C) or (B, T, H, W, C) form - if there's a batch dimension it just takes the first batch entry, so you can feed it a batched set without thinking. Then it's a plain index:

• first_frame = true → take frame at offset from the start.
• first_frame = false → take frame at offset from the end.

The offset gets clamped into range rather than erroring, so asking for offset 5 on a 4-frame clip silently gives you frame 0 instead of crashing. You get one image output with a fresh batch dimension, ready to drop into anything that expects an IMAGE.

The inputs that matter

There are only three, and you'll actually touch two:

• video - the clip or image sequence.
• first_frame - which end you're counting from. Default true.
• offset - how many frames in from that end. Default 0, which is "give me the very first (or very last) frame."

Installing it

Same routine as every node in the pack - ComfyUI Manager, search "ComfyUI-SA-Nodes-QQ", or:

cd ComfyUI/custom_nodes
git clone https://github.com/siraxe/ComfyUI-WanVideoWrapper_QQ.git

Restart after cloning. No extra dependencies, no model downloads for this node.

Where people get burned

Honestly, the trap here is expecting more than it does. If you want the average frame, or a frame at a percentage position in the middle, or the first frame of a specific batch entry - that's not this node. offset is a count, not a fraction, and first_frame=false with offset=0 gives you the last frame, which people routinely misread. That's also the combo you want for the chunk-chaining use case, so remember it. Beyond that it's about as frictionless as a ComfyUI node gets: feed it frames, get one frame.