Kling IMAGE 3.0

Kling IMAGE 3.0 is Kuaishou's text-to-image model - the same house that made Kling video - and this node drops it into ComfyUI with a single required input and no GPU of your own involved. Unlike the video nodes, the output is a plain IMAGE tensor, so it plugs straight into the ComfyUI image world you already know: preview, save, pass it along as the first frame of a Kling video node, whatever you like.

Why you'd reach for it

The closed-image-model problem, solved the same way the pack solves everything: run it on Runware's cloud. Kling IMAGE is API-only, no weights, and its quality is genuinely competitive with the big cloud image models. The neat move is workflow continuity - you can generate here, then wire the IMAGE output directly into a Kling video node's frameImages and animate it, all in one graph. That's the pipeline this node exists for.

How it works

Standard generated-node machinery: positivePrompt is required, and any optional referenceImages tensor is base64'd to PNG data URIs. The request rides to the Runware REST API via the runware SDK and returns a native IMAGE tensor (not a file path - a real ComfyUI image). Cost and NSFW flag print on the node title bar. Image nodes are also the cheap end of Runware's pricing, so this is a friendly way to test the pack.

Inputs that matter

• positivePrompt - required.
• referenceImages (IMAGE) - optional. Feed one or more images and you're doing image-guided generation (reference, style, edit) instead of pure T2I.
• size - the resolution picker: presets from 1k/2k (from input) when you've wired references, through the whole 1K/2K aspect-ratio grid (1:1, 3:2, 2:3, 4:3, 9:16, 16:9, 21:9).
• numberResults (1–20!) - this one rolls up to 20 variations per run, which is generous for a paid API; remember each one costs.
• outputFormat (JPG/PNG/WEBP), outputQuality (95).
• watermark - an optional typed watermark input if you want one stamped in.
• safety group - content-check toggles (off = model default).

Output: one image socket (IMAGE type).

Install and the API key

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart ComfyUI (or Runware via ComfyUI Manager). Key from the Runware dashboard → Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Gotchas

• referenceImages wants tensors from Load Image - a file path or URL string will error.
• Watch numberResults: 20 variations is a real bill. It's there for batch work, not for curiosity.
• Closed model, per-run pricing: Kuaishou's weights rented by the request, cost shown on the node.
• size and any width/height-style inputs are mutually exclusive - the dropdown owns the resolution, so set it there, not on hidden widgets.

Kling IMAGE 3.0 is the gateway node of the pack: cheap, fast, and the natural first frame for everything else Kling does. If you're going to animate a Kling video, starting here is the intended loop.