KlingAI 1.0 Pro

Kling 1.0 is the model that taught people frontier video had to be bought, not built. It's Kuaishou's closed video model - no weights, no local checkpoint, API-only - and this node is the way to drive it from your ComfyUI graph without ever leaving the canvas. Before you ask: yes, this one really does call an API, and yes, it really does need a key and a few credits. It's the opposite of the local nodes you're used to.

Why you'd reach for it

If you're running Wan or Hunyuan locally, you know the deal: great results, but a 16GB+ card, long queues, and motion quality that still loses to the big cloud models. Kling was the quality bar the open-source world spent 2024–2025 chasing. This node gives you that ceiling with zero local inference - your prompt is shipped off to Runware's cloud, a GPU you don't own renders the clip, and a finished MP4 comes back. If your machine can't run a video model at all, this is one of the cheapest doors into Kling-quality video from inside ComfyUI.

How it works

This is a Runware cloud node, generated straight from the company's model catalog. The node class is thin - all the logic lives in the pack's base class. On queue, it bundles your prompt and any image inputs into a request, base64s your images to PNG data URIs, POSTs to the Runware REST API through the runware SDK, then downloads the result and wraps it as a native ComfyUI VIDEO object. Each run shows its cost on the node title bar, which is a nice honesty feature.

Inputs that matter

Only positivePrompt is required. The rest are optional, and honestly you'll mostly touch a handful:

• frameImages (IMAGE) - feed this a tensor and your text-to-video becomes image-to-video. Leave it empty for pure T2V.
• duration - 5 or 10 seconds. Longer costs more.
• width / height - default 1024×1024; set your aspect.
• CFGScale - defaults to 0.5. With guidance that low, negativePrompt is ignored, so don't bother writing one unless you raise it.
• numberResults (1–4) - variations per run, each with its own seed.
• outputFormat - MP4, WEBM, or MOV. outputQuality is the encoder quality (95 default).

The single output is video (type VIDEO), which you wire into a video preview or save node.

Install

Install the pack once and every Kling node is yours:

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

Restart ComfyUI. Easier still: ComfyUI Manager → search Runware → install. Then grab a key from the Runware dashboard and either paste it into Settings → Runware API key, export RUNWARE_API_KEY, or run runware auth login once.

Gotchas

• No key means no run: you'll get a clean one-line "No Runware API key" error. Set RUNWARE_DEBUG=1 if you ever need the full traceback.
• The VIDEO output type needs a reasonably current ComfyUI; on older builds the pack degrades to returning a saved file path instead.
• You're paying per generation. Runware is cheap per image, but video adds up fast - a bad prompt is a few cents you can't get back, so set numberResults low while you iterate.
• Kling is Kuaishou's, not Google's or Bytedance's, and it's firmly closed-source. If you want something you can tinker with locally, this isn't it. If you want Kling's look on a machine that can't run a video model, it's exactly it.

The 1.0 line is old news in the Kling world by now - 1.6, 2.x, and the video 3.0 nodes all outclass it. Grab it if you're price-comparing old generations, or if a shared workflow pinned to 1.0 won't budge.