Cloud LTXV Latent Upsampler

There are two ways to make a low-res video bigger: decode it and run a pixel upscaler, or upscale the latent before decoding, using a dedicated latent upscale model that knows what latent space actually looks like. Cloud LTXV Latent Upsampler is the second approach, contributed as an LTXVLatentUpsampler step in the cloud workflow JSON. For LTX-2 - a model that already renders internally at half resolution before its own upscale - latent upscaling is often the cheaper, more coherent path than a pixel pass.

It's the cloud mirror of the LTX latent upsampler you'd drop into a local LTX-2 template between sampling and decode. The upscale model runs cloud-side, so you never download it; your job is just to wire the three inputs together correctly.

Inputs

• samples - the CLOUD_LATENT from your sampler, at its native (lower) resolution.
• upscale_model - a CLOUD_LATENT_UPSCALE_MODEL handle. You don't load this directly; the pack ships a Cloud Latent Upscale Model Loader node for exactly this handle type, and it has its own dropdown cache. Load the upscale model first, wire it here.
• vae - a CLOUD_VAE handle. The upsampler needs to round-trip through the VAE as part of its job, so the VAE has to be in the chain too.

Output is one LATENT (CLOUD_LATENT) at the upscaled resolution, which goes into your VAE decode.

Installation

Standard pack install:

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

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

Gotchas

• You need the loader node. upscale_model isn't a dropdown on this node - the load happens in the separate Cloud Latent Upscale Model Loader, which is where the model dropdown lives and where the usual "run Cloud List Models + restart" refresh dance applies.
• Latent upscaling is a different beast from pixel upscaling (the KB's upscaling coverage treats them as separate tools). If you want to apply a local latent upscaler that isn't in the cloud's catalog, the pack's Cloud Materialize Latent node is the bridge that drags the latent back to your GPU for exactly that.
• Whole-pack status check: proof of concept from a small community project, so the upscale-model options on Comfy Cloud are whatever happens to be there - if the loader dropdown is thin, that's the cloud's catalog, not a bug in this node.