Kling IMAGE O1

The tell is in the size dropdown. Kling IMAGE O1 offers 1k (from input) and 2k (from input) as its first options - this is a model designed around images you bring, not just words you type. It's Kuaishou's O-series image model, API-only, and this node puts it in ComfyUI where the output comes back as a plain IMAGE tensor you can preview, save, or feed forward into the rest of your graph.

Why you'd reach for it

If you live in the generate-a-frame, then-animate-it loop, O1 is built for the reference side of that loop. The referenceImages socket and the "from input" size presets are the model telling you how it wants to work: give it a source, let it produce at the source's resolution. Compared to the video nodes, image generation on Runware is also cheap enough that you can actually experiment here.

How it works

Generated Runware node, shared base class. Your optional referenceImages tensors are base64'd to PNG data URIs and sent with the required positivePrompt to the Runware REST API over the runware SDK; the result returns as a native IMAGE tensor. The size dropdown folds resolution into one choice - when set to a from input preset, the output matches your reference; when set to a 1K/2K preset, it renders at that aspect ratio. Cost and NSFW flag print on the title bar.

Inputs that matter

• positivePrompt - required.
• referenceImages (IMAGE) - optional, but this model's whole point. Feed images for reference/guided generation.
• size - 1k (from input), 2k (from input), or the 1K/2K aspect grid (1:1, 3:2, 2:3, 4:3, 3:4, 9:16, 16:9, 21:9).
• numberResults (1–20) - variations per run; each one costs.
• outputFormat (JPG/PNG/WEBP), outputQuality (95).
• watermark - optional typed watermark input.
• safety group - content-check toggles.

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 is an IMAGE tensor - Load Image first, not a path or URL.
• "From input" sizes only make sense when you've actually wired a reference; without one, pick a concrete preset.
• numberResults up to 20 is a serious bill if you let it run wide. Batch thoughtfully.
• Closed Kuaishou model, per-request pricing, cost on the node. You're renting the weights, not owning them.

O1 is the "start from something you already have" node in Kling's image line. If your workflow is reference-in, generation-out, this is the one that matches how you think.