Cloud VAE Loader
A whole node for one dropdown — but when your model has no bundled VAE, it's the only way
- vae
Cloud VAE Loader looks comically small: one dropdown, one output. But it exists for a structural reason. In a checkpoint workflow the VAE comes along for the ride, baked into the checkpoint file. In a diffusion-model workflow - which is what this pack is mostly about, since video models like Wan and LTX ship as standalone diffusion files - there's no bundled VAE. Somebody has to pick one. This node is that somebody.
How it works
Same pack machinery as the other loaders: it takes your pick from the cloud's vae folder and appends a VAELoader node to the assembled workflow JSON. Nothing loads locally; the load happens on the cloud instance when a terminal node finally fires. What you get back is a CLOUD_VAE handle that threads into the encode/decode steps downstream.
The inputs and outputs that matter
- vae_name - dropdown populated from the cloud's
vaefolder. It shows a placeholder (<run Cloud List Models (vae) to populate>) until the cache is filled. - Output: vae (CLOUD_VAE) - wire it into Cloud VAE Decode (latents → pixels) or Cloud VAE Encode (local image → latents).
When you need it
Load your diffusion model with Cloud UNET Loader and you have no VAE; you need this node before any decode or encode step. Even in a checkpoint-based workflow you might reach for a standalone VAE if the bundled one is known-weak - this is the same reflex that makes people swap VAEs in local ComfyUI. And Wan specifically uses its own VAE file, so I2V pipelines on this pack (like the bundled video_wan2_2_14B_i2v_cloud.json example) always run through a Cloud VAE Loader.
Installing it
Install via ComfyUI Manager (search "ComfyUI-CloudAPI-worker") or:
cd ComfyUI/custom_nodes
git clone https://github.com/Dobidop/ComfyUI-CloudAPI-worker
cd ComfyUI-CloudAPI-worker
cp config.json.example config.json
Paste your API key from https://platform.comfy.org/profile/api-keys into config.json and restart ComfyUI. Only requests, Pillow and safetensors locally - the VAE itself never downloads to your machine.
Common issues & troubleshooting
- The classic: placeholder dropdown - run a Cloud List Models node with folder=
vae, then restart ComfyUI. The startup prefetch refreshes on a 24-hour cadence, so if you added a new VAE cloud-side you can either wait or force it with Cloud List Models. - "No VAE selected" error - that's the guard firing when
vae_namestarts with<. The message tells you exactly what to do; it's a state problem (empty cache), not a code problem. - Different VAE than expected in results - make sure the VAE you pick matches the model family. Wan wants its own VAE; an SD1.5-era VAE picked from the same dropdown will decode garbage. The dropdown lists names, not families, so read before you click.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| vae_name | COMBO | 1 options: <run Cloud List Models (vae) to populate> |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| vae | CLOUD_VAE | — |