KlingAI 2.0 Master

Kling 2.0 Master is the model that set the bar the local video community spent the next year trying to reach. When the Wan crowd benchmarked their best open-weight work, the honest framing was "on par with Kling 2.0 Master" - that's the yardstick. This node is Kuaishou's flagship tier of the 2.0 generation, running on Runware's cloud and driven from a ComfyUI node that needs exactly one thing from you: a prompt.

Why you'd reach for it

Because this is where "I'll just run it locally" stops being an option. 2.0 Master has no weights you can download - Kling is closed, API-only, and the Master tier is the top rung. The Runware node is your front door: positivePrompt is the only required input, and the render happens on Runware's GPUs and comes back as a native ComfyUI VIDEO. If you've been fighting Wan's motion quality and wondering what the actual ceiling looks like, this is the answer - rented, not owned.

How it works

Standard pack mechanics, flagship model. The node class is generated from Runware's catalog with the 2.0 Master model ID baked in; your prompt and any optional frameImages tensor (base64'd to PNG data URIs) get sent to the Runware REST API via the runware SDK, and the finished clip is downloaded and wrapped as a VIDEO object. The node title bar shows the cost per run and any NSFW flag. No local inference at all.

Inputs that matter

• positivePrompt - required. This is a T2V-first node; describe the whole scene.
• frameImages (IMAGE) - optional. Feed a first frame and you get image-to-video on the same model.
• duration - 5 or 10 seconds. Longer clips on a flagship tier are where credits go to die; budget accordingly.
• width / height - 1024×1024 default. Master tier handles it well, but resolution costs money on every frame.
• CFGScale - 0.5 default; negativePrompt ignored below guidance ~1.
• numberResults (1–4) - variations per run, each a separate bill.
• outputFormat (MP4/WEBM/MOV), outputQuality (95).

Output: one video socket (VIDEO type) → preview/save node.

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

• It's the most expensive Kling tier in the pack. The title-bar cost line will make this obvious fast - set numberResults to 1 and nail the prompt before you roll for variations.
• No key = clean one-line "No Runware API key" error; RUNWARE_DEBUG=1 for full tracebacks.
• frameImages is an IMAGE tensor (Load Image), never a path.
• Closed weights, per-request pricing: you own the output, not the model.

2.0 Master is the one you reach for when the shot has to be right and the local alternatives just aren't there yet. It's also the one that reminds you cloud inference is a subscription, not a purchase - worth it exactly as often as you're willing to pay for the ceiling.