Cloud Empty Image

Cloud Empty Image creates a blank image - in the cloud. It's the pixel-space counterpart to Cloud Empty Latent: same idea, but instead of producing a CLOUD_LATENT for text-to-image chains, it produces a CLOUD_IMAGE handle backed by an EmptyImage node that will exist only when your workflow actually runs on Comfy Cloud.

Why would you want a blank image? A few genuinely useful reasons. It's the seed canvas for image-space chains that need a starting frame - the pack's image utilities (Cloud Image Scale By, Cloud Get Image Size, Cloud Resize Images By Longer Edge) all eat CLOUD_IMAGE handles, and an empty one is a fine way to drive size math. It's also the escape hatch when a graph needs an image-shaped input but you don't have a real one yet. And because color lets you pick anything from black to white to neon, it doubles as a cheap solid-color test card for checking that a chain assembles correctly before you burn cloud credits on a real generation.

The inputs that matter

• width / height - default 512×512, up to 16384. Same multiples-of-8 habits as any latent sizing apply here.
• batch_size - default 1, up to 4096. This is how you get N blank frames cheaply for a batch test.
• color - an integer 0–16777215, which is just RGB in disguise (0x000000 = black, 0xFFFFFF = white). The ComfyUI UI surfaces it as a color picker; the value is passed straight through to the cloud's EmptyImage node.

Output: image (CLOUD_IMAGE) - feed it to Cloud Image Scale By, Cloud Get Image Size, or a cloud VAE-encode-style chain. It's a handle, not pixels: you can't preview it locally until a terminal node (Cloud Fetch Images, Cloud VAE Decode) actually runs the chain and downloads the result.

How it works

The whole pack runs on handles - lightweight references that accumulate a workflow JSON, never actual data. Cloud Empty Image just appends an EmptyImage spec with your four values to that dict. Nothing renders at graph-build time, nothing is previewed, nothing costs a cent until you hit a terminal node and the assembled workflow gets POSTed to /api/prompt.

How to 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. Or install via ComfyUI Manager. Dependencies: requests, Pillow, safetensors. No model downloads - the blank image is generated cloud-side.

Common issues

• "I see nothing" - expected. Handles aren't previewable; wire the chain to a terminal before you judge.
• Forgetting this isn't a real canvas - a blank CLOUD_IMAGE is not a local image. If your intent was img2img from a real picture, that's what Cloud Upload Image (or the optional image input on the Quick sampler) is for.
• PoC expectations - this is a utility node in a pack the author calls "very hacked together." It does its one job; just don't expect it to do more than mirror the stock EmptyImage.