API Core

API Core is the pack's take on "generate on someone else's GPU." It's a hosted image-generation API call wrapped as a node: you type a prompt, paste an API key, and the response comes back as image tensors you can preview, save, or feed into the rest of your workflow. The default model is gemini-3-pro-image-preview and the default prompt is, no joke, "画个类似图片" - "draw a similar image" - because the interesting mode here isn't text-to-image, it's image-to-image with a reference.

The service behind it is API Core (apicore.ai), a third-party API broker. The README points you to their console to apply for a token, and it's a paid service - this is not a free ride, and it needs internet. If your workflow is built to run entirely offline, this node and its cousins in the pack are the exception, not the rule.

How it works

The flow is: you give it a prompt, an api_key, and a model. The moment you fill in a valid key, the node pulls the list of available models from the API so you can pick from what's actually live rather than guessing. Optional image_1 input slots (more appear as you connect them) let you pass reference images, which is where the "similar image" default prompt comes from - the intended use is generating variants of an image you already have.

There's one honest gotcha here: to send your reference image to a third-party server, the node has to give that server a URL to fetch. It builds one from your ComfyUI address - the server_origin field - which means your ComfyUI must be reachable at a public IP or domain. A localhost-only box can't be fetched by an external API. This trips up more people than any other setting on this node.

The inputs that matter

• server_origin - required, and it has to be a public address of your ComfyUI, e.g. http://192.168.1.10:8188 won't work if that IP isn't reachable from the internet.
• api_key - from the API Core console.
• size - ten aspect ratios (1:1, 2:3, 9:16, 21:9, etc.) for the output canvas.
• n - how many images to generate, 1 to 10.

Outputs: response (the raw API response as text) plus preview_1 through preview_10 as IMAGE tensors. It's flagged as an output node, so it's meant to sit at the end of a branch and render.

Install

Ships in ComfyUI-WBLESS. No requirements.txt, no models to download:

cd ComfyUI/custom_nodes
git clone https://github.com/LaoMaoBoss/ComfyUI-WBLESS

restart ComfyUI. Or ComfyUI Manager → search "ComfyUI-WBLESS" → install.

Troubleshooting

If reference images silently fail, check server_origin first - this is the number-one failure mode, and it's usually a router or firewall problem, not a node problem. If the model dropdown stays empty, your key is wrong or the service hasn't authorized it yet. And keep expectations realistic: you're at the mercy of a third-party API for availability, pricing, and whether the model you want is even offered that week. When it works it's great - a hosted model you can't run locally, dropping images straight into your graph. When the API is down, it's just a node that's waiting on someone else's server.