IAMCCS Cine Filmmaker Crop Guides 1:1

Here's a thing nobody warns you about with guide-based video generation: when you inject reference keyframes into an LTX latent, those guide frames can get appended to the latent during sampling, and they're still there when you decode. Your 8-second clip comes out with an extra few frames of guide garbage stuck on the end, or worse, your VAE decode chokes on a latent that's longer than your timeline. IAMCCS Cine Filmmaker Crop Guides 1:1 is the cleanup node that trims those appended guide frames off, using the guide contract that the conditioning itself carries.

It's a small, focused utility from the IAMCCS Cine pack, and it's specific to the "1:1" single-shot generation path (the author's naming: guide handling that matches the LTX Director Guide behavior one-to-one). The "WhatDreamsCost contract" in the source docstring is shorthand for how the guide tail is tracked - the node reads how many guide frames were added from the conditioning and crops the latent accordingly.

How it works

You pass through positive, negative, and latent - the same objects the guide nodes gave you. It looks at the conditioning to figure out how many guide-crop frames are appended (_guide_crop_count_from_conditioning), then crops the latent and the noise mask down to the real shot length. It passes positive/negative straight through untouched and returns the trimmed latent, ready for VAE decode.

Inputs and outputs

Only three inputs - positive, negative, latent - and the same three come back out. No widgets, no knobs, no guessing. It belongs right before your VAE decode, after the FLF Productor / guide application and sampling.

Installing it

Part of IAMCCS-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart ComfyUI (or Manager → search "IAMCCS"). Baseline: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8. No models or extra deps.

The honest caveat

This is a "you know you need it when you need it" node, and you'll know because your decoded video is the wrong length or has junk frames at the end. If you're running a plain shot with no reference guides, it does nothing useful (it crops zero frames) - you don't need it on every graph. And it's the 1:1-specific variant: if your workflow uses the Cine system's own FLF path rather than the LTX Director-compatible guide contract, the Cine FLF Length Compensator is the one that does the pre-computation for that path. Matching the right cleanup node to your guide path is half the battle with this pack.