Nodes/IAMCCS-nodes/IAMCCS LTX Video Duration Crop
ComfyUI Node

IAMCCS LTX Video Duration Crop

Snip the corrupt tail frames LTX-2 adds to every clip

By IAMCCS·Created 11 months ago·Updated 8 days ago· 113
IAMCCS LTX Video Duration Crop
  • video
  • guide_data
  • video
manual_duration_frames0

If you've rendered LTX-2 clips longer than a few seconds, you've seen it: the video looks right, then the last few frames go soft, noisy, or visibly corrupt. That's not a model failure or a seed problem. LTX-2's VAE only encodes in 8n + 1 frame counts, so when your pipeline asked for, say, 80 frames, the sampler quietly produced 81 - and the frames past your real target were generated outside the conditioning window. IAMCCS_LTXVideoDurationCrop exists to cut those strays off before you save.

It's part of the IAMCCS-nodes Shotboard V4 ecosystem - the production-oriented video tooling from IAMCCS. The node's own docstring says exactly where it belongs: after VAE decode / VideoCombine and before SaveVideo or the Shotboard video editor.

How it works

The node needs to know how many frames you actually wanted. That value lives in guide_data["duration_frames"] - the pre-rounding target that the IAMCCS CineFilmmakerBackend (V3) and CineShotboardV4Backend (V4) write into their GUIDE_DATA hand-off. Give it that, and it crops the incoming VIDEO back down to the true duration, dropping the padded, un-conditioned tail frames that the 8n + 1 rounding appended.

If you're not running the full Shotboard pipeline, manual_duration_frames is the escape hatch: set it above 0 and the node trims to that exact count, ignoring guide_data entirely. Manual wins when both are present.

Inputs and outputs

  • video (VIDEO) - required. The decoded video from your VAE-decode / VideoCombine chain.
  • guide_data (GUIDE_DATA) - optional. The author's own data contract; the node reads duration_frames from it.
  • manual_duration_frames (INT, default 0) - manual override. If > 0, trims to this exact frame count.

Output:

  • video (VIDEO) - the trimmed clip, ready for SaveVideo or the editor.

No knobs beyond that. It's a crop, done at the VIDEO level so it slots into the modern ComfyUI video pipeline rather than the old image-batch world.

How to install

Comes with IAMCCS-nodes - ComfyUI Manager search "IAMCCS", or:

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

Restart, and the node appears under IAMCCS/Cine/Shotboard V4. No extra model downloads; it's pure video-plumbing code. Pack requirements are ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Gotchas

The node passes video through unchanged if it can't find a target - either no guide_data and manual_duration_frames left at 0, or the comfy_api.latest import failing on an older ComfyUI. So if you wire it up and the tail is still there, check you actually connected guide_data (or set the manual count) rather than assuming it silently guessed. And no: this doesn't fix the start of your clip. If you're getting glitchy fast motion in the opening frames, that's LTX-2's I2V behavior - the community fix is raising the FPS for I2V, not cropping.

CategoryIAMCCS/Cine/Shotboard V4

Inputs (3)

NameTypeDefaultDescription
videoVIDEO
guide_dataoptGUIDE_DATA
manual_duration_framesoptINT00–100000Manual override: if > 0, trim to this exact frame count. If 0, reads automatically from guide_data['duration_frames'].

Outputs (1)

NameTypeDescription
videoVIDEO