Cloud LTXV Crop Guides

Some LTX workflows feed the model "guide" frames - reference frames baked into the latent to steer the video. The problem: those guides live inside the latent you sampled, and if you don't cut them out before decoding, they show up in the final clip as unwanted duplicated frames. Cloud LTXV Crop Guides is the node that trims them, contributing an LTXVCropGuides step to the cloud workflow JSON.

The source docstring calls it exactly that: "Crops conditioning guide frames out of an LTX latent." It's the return-trip bookend to the guide-injection nodes - you bake guides in at the start (via something like the img2video or guide-add path), and this cuts them out after sampling. If you're new to LTX, treat this as a workflow-template node: it's in the reference graph, it does its quiet cleanup job, and the reason it's there becomes obvious the first time you forget it and get a clip that starts with a frozen duplicate of your input frame.

Inputs

• positive and negative - CLOUD_CONDITIONING handles.
• latent - the CLOUD_LATENT straight out of your sampler.

Outputs: fresh positive, negative, and a cropped latent. The cropped latent is the one that goes into the VAE decode; the conditioning comes back because the crop step also adjusts the guide references inside the conditioning.

Installation

Same pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Dobidop/ComfyUI-CloudAPI-worker

Copy config.json.example to config.json, add an API key from https://platform.comfy.org/profile/api-keys, restart. In ComfyUI Manager under "ComfyUI-CloudAPI-worker"; deps are requests, Pillow, safetensors.

Gotchas

• It only matters if your workflow injects guides. For a plain text-to-video LTX chain with no guide frames, this node is a no-op passthrough - harmless, but skip it if you want a leaner graph.
• All three outputs need to flow forward. If you crop the latent but still feed the old conditioning to the next step, the guide bookkeeping gets confused. Route the node's own outputs everywhere.
• The pack is a proof of concept; the LTX nodes were added in an update the author described as working "well with the example workflow at least." When in doubt, mirror example_LTX_workflow.json from the repo.