Cloud LTXV Preprocess

LTX image-to-video is picky about its input images in ways that don't show up in text-to-image. Feed it a 4K screenshot and the model spends its first frames burning compute on detail it's going to regenerate anyway - and you pay for that on Comfy Cloud by the job. Cloud LTXV Preprocess is the fix: it compresses the start image cloud-side before it goes into the I2V path, contributing an LTXVPreprocess step to the assembled workflow JSON.

The source docstring is honest about what it does: "Compresses an image (cloud-side) before feeding into LTX I2V." It's a one-trick node, and the trick is exactly the kind of thing that matters more in a metered cloud context than locally - every wasted pixel is a job that runs a little longer.

Inputs

• image - a CLOUD_IMAGE handle (from an uploaded or generated image; local pixels cross into cloud-land through the pack's upload bridge).
• img_compression - INT, default 35, range 0–100. How aggressively to compress. Higher = smaller/faster but more quality loss in the start frame. 35 is a sane default; crank it up for drafts, down for final quality.

Output: one output_image (CLOUD_IMAGE), the compressed image ready for Cloud LTXV Img To Video Inplace.

Installation

Same pack as always:

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. ComfyUI Manager has it under "ComfyUI-CloudAPI-worker"; deps are just requests, Pillow, safetensors.

Gotchas

• Don't confuse this with resizing. Cloud Resize Images By Longer Edge fixes the dimensions; this fixes the compression. If your input is huge, you probably want both - resize to the model's preferred resolution, then compress.
• img_compression at 0 means "don't compress," which is fine for a small input and wasteful for a big one. There's no right answer, just a cost/quality slider.
• Proof-of-concept pack. The LTX path was the author's second update, tested mainly against the included example workflow - if an I2V run looks off, check whether you even needed this node in the chain or whether your start image was already reasonable.