Drop First Frames

Your Wan I2V clip opens with garbage. Not always, but often enough that you learn to expect it: a flash, a smear, a few frames of the model treating your carefully composed starting image as a polite suggestion. The author of this pack knows - they even warn about it in their own docs. Drop First Frames is the boring, reliable fix.

It's a tiny utility node that ships inside comfyui-wan-i2v-control, and honestly it's worth having even if you never touch the pack's main masking node. It does exactly one thing: chops the first N frames off an image batch. And because it works on any IMAGE batch, not just Wan output, you'll find yourself reaching for it in all kinds of pipelines where a sequence starts with junk.

How it works

Nothing clever under the hood, which is the point. An image batch in ComfyUI is a tensor shaped [B, H, W, C] where B is the frame count. The node slices off the leading frames and returns the rest:

• images - the batch of frames (required)
• frames_to_drop - integer, default 4, range 0–32. 0 is a passthrough (handy as a disable switch).

One nice safety touch in the code: if you ask it to drop more frames than actually exist, it returns the last frame rather than erroring out or handing you an empty batch. No crash, just a very short clip.

Output is a single images - the same type you fed in, just shorter. Wire it between your sampler output and whatever encodes or saves the video (VHS VideoCombine, SaveAnimatedWEBP, and so on), and you've cut the glitch frames before they ever reach a file.

Installing it

It comes with the pack, so install is the shared story:

cd ComfyUI/custom_nodes
git clone https://github.com/shootthesound/comfyui-wan-i2v-control
pip install mediapipe

Restart ComfyUI. You can also find it via ComfyUI Manager by searching "WAN I2V Control". The only real dependency the pack declares is mediapipe (that's for the masking node's person detection - Drop First Frames itself just needs PyTorch, which you already have). There are no model files to download for this node.

Common issues

This one's hard to break, but a couple of things to know:

• Drop too many frames and the clip shrinks. It doesn't regenerate anything - the output is simply N frames shorter. If you need a specific duration after trimming, you'll have to pad or re-time downstream.
• Still seeing a bad frame after the default of 4? Bump frames_to_drop up. Some Wan generations are glitchier than others, especially at higher resolutions or with aggressive motion.
• A dead-end on the first frame is a different problem. If the entire clip is wrong, trimming won't save you - that's a prompt/conditioning issue, not a frame-glitch issue. This node only cuts off the leading garbage.

For such a small node it's oddly easy to take for granted - until you watch one clip without it and remember why it exists.