Cloud LTXV Img To Video Inplace

Image-to-video in LTX works by taking your start image, encoding it, and baking it directly into the latent - the first frames of the video already contain your image, and the model animates out from there. That's the "inplace" in this node's name: it contributes an LTXVImgToVideoInplace step to the cloud workflow JSON that fuses the image into the latent rather than adding it as a separate conditioning branch.

Where does this fit in a cloud chain? You've got a CLOUD_IMAGE (from an uploaded or generated image), a CLOUD_LATENT (the empty video latent), and a CLOUD_VAE to encode with. This node takes all three, returns a single latent that's now "primed" with your starting frame, and that latent goes to the sampler. It's the LTX-specific way of doing I2V, distinct from the Wan I2V path the pack also supports.

Inputs

• vae - a CLOUD_VAE handle, the image VAE that encodes your start frame.
• image - a CLOUD_IMAGE handle. Remember the pack's rule: local images are uploaded once via /api/upload/image and become a LoadImage node in the assembled workflow, so your local LoadImage feeds a bridge that turns it into CLOUD_IMAGE.
• latent - the CLOUD_LATENT you're priming, from Cloud Empty Latent Video.
• strength - FLOAT, default 1, range 0–1. How much of the image survives into the video. Lower values let the model diverge from the input more freely.
• bypass - BOOLEAN, default false. Handy switch: flip it true to run the chain as plain text-to-video without rewiring the graph.

Output: one latent (CLOUD_LATENT).

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 by pack name; deps are requests, Pillow, safetensors.

Gotchas

• The image has to actually reach the cloud. If you wired a local IMAGE straight into this node you'll hit a type mismatch - the bridge from local pixels to CLOUD_IMAGE happens in the pack's upload node, not here.
• strength near 1 can make the video hug the input so tightly it barely moves; near 0 and you've basically defeated the point of I2V. Start at 1 and back it off if the animation feels stiff.
• Proof-of-concept pack; the LTX nodes were the author's follow-up update, described as working "well with the example workflow at least." The repo's example workflows are the closest thing to documentation, so mirror them.