Cloud CLIP Text Encode

Cloud CLIP Text Encode is the prompt box of a cloud workflow. It takes a CLOUD_CLIP handle and your text and returns CLOUD_CONDITIONING, exactly like the stock CLIPTextEncode - except the encoding happens on Comfy Cloud's hardware, not yours. You'll normally drop two of these in a graph: one for the positive prompt, one for the negative, both wired from the clip output of a loader.

Mechanically it's the same trick as every other node in this pack. The node appends a CLIPTextEncode spec to the accumulating workflow JSON - {"class_type": "CLIPTextEncode", "inputs": {"clip": [...ref...], "text": "your prompt"}} - and hands you a handle referencing it. Nothing is encoded when you press run. Only the terminal node (Cloud Fetch Images, Cloud VAE Decode, Cloud Save Video, or the Quick sampler) actually submits the whole graph to /api/prompt and downloads the result.

The inputs that matter

• clip (CLOUD_CLIP) - from a Cloud CLIP Loader, Cloud Dual CLIP Loader, or the clip output of a Cloud Checkpoint Loader.
• text - multiline string, your prompt. Empty by default, so the "why is my negative doing nothing" failure mode here is you simply forgetting to type anything.

Output: conditioning (CLOUD_CONDITIONING) - feed it into Cloud KSampler Graph, Cloud KSampler Advanced, or Cloud CFG Guider.

What you give up

Because the conditioning is computed cloud-side by the stock CLIPTextEncode, the local conveniences you might be used to won't carry over. Conditioning combos, regional prompts, or any other pack that rewrites the conditioning object between encode and sample won't have a cloud counterpart here. The pack's converter (Cloud Convert Workflow) deliberately leaves boundary nodes alone and only swaps class names for the cloud equivalents it knows. If your local graph depends on exotic conditioning plumbing, budget for it not surviving the trip.

Also worth stating plainly: the text you type goes over the network as part of the job payload. That's the nature of every API-wrapper node - the same reason the KB's guidance on this category is "prefer widely-installed open packs and read what a fresh one does." For a personal prompt this is usually fine; for client work or anything sensitive, think twice about what you type into a cloud chain.

How to install

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

Then copy config.json.example to config.json, paste an API key from https://platform.comfy.org/profile/api-keys, and restart ComfyUI (or install via ComfyUI Manager). Dependencies: requests, Pillow, safetensors - nothing heavy, no models.

Common issues

• Prompt silently not doing anything - check you actually typed text, and that the clip input is a cloud handle from a cloud loader, not a local one. Mixed chains won't connect.
• No output at all - again, this node is mid-chain. It never submits anything on its own; you need a terminal downstream.
• PoC caveat - the pack works well for the example image and LTX-video workflows, and the author calls it "very hacked together" everywhere else. Start from the repo's example workflows before building something custom.