Skip to main content
The Video Generator node runs an AI video model and produces a video as output. You can drive it with a text prompt, an image-to-video workflow where a reference image is animated, or a full pipeline where upstream nodes produce both the prompt and the starting frame automatically.

​
Available models

The model picker lists every enabled video model in your workspace. Seven models are available by default, organized into two model families: Seedance 2.0 family — high-quality text-to-video and image-to-video, multi-mode (UGC, lipsync, first-last-frame, etc.)
  • Seedance 2.0 Full Access — the full-featured variant with every mode unlocked
  • Seedance 2 Fast — optimized for speed
  • Seedance 2.0 — the base variant
Kling V3 / V2.5 family — image-to-video and motion control:
  • Kling V3 Pro I2V — image-to-video (supports an advanced 4K mode via API parameter; see callout below)
  • Kling Video v2.5 Turbo Pro — fast turbo variant
  • Kling V3 Motion — motion control
  • Kling O3 Pro Ref2V — reference-to-video with style + character refs
Seedance 2.0 variants require a Business tier subscription. Workflows that include any Seedance 2.0 node show a lock badge on the template card if your account is on a lower tier. Kling models are available on Pro+.
Kling V3 Pro 4K mode — API-only. Kling V3 Pro accepts an optional 4k boolean parameter that requests a 4K rendering pass instead of the default 1080p. 4K is API/MCP/CLI-only — it is intentionally not exposed on the canvas because 4K renders typically take 15–25 minutes per 5-second clip, which is not a workable canvas UX. Output file sizes roughly double vs 1080p, and cost scales accordingly.To use 4K, call the model via the slim API with 4k: true in the input:
POST /api/models/kling-video-v3-pro-image-to-video/run
{
  "input": {
    "start_image_url": "https://...",
    "prompt": "your prompt",
    "duration": "5",
    "4k": true
  },
  "async": true
}
Without the 4k parameter (or with 4k: false), the request renders at the standard 1080p resolution. async: true is strongly recommended for 4K so your client does not hold an HTTP connection open for the full render window — Knouds returns an execution id immediately and you poll via GET /api/executions/:id.
Your workspace administrator can add more video models through the Node Builder.

​
Input sockets

The sockets available depend on the selected model. Common inputs include:
SocketTypeDescription
prompt-inText (orange)The generation prompt
image-inImage (pink)Starting reference image for image-to-video
end-image-inImage (pink)Ending frame for morphing workflows
audio-inAudio (cyan)Audio track to incorporate into the video
params-inAny (purple)Parameter bundle from a Request Input node
Switching models updates the socket set. The node attempts to reconnect existing wires to compatible handles automatically.

​
Output sockets

The video-out socket on the right side emits the generated video. Connect it to a Response Output to return it to an API caller, or to a Frame Extractor node to continue processing frames.

​
Controls on the node

First row
  • Model picker — select the active model.
  • Aspect ratio dropdown — available options depend on the model (for example, 16:9, 9:16, 1:1).
Second row
  • Duration dropdown — available durations depend on the model (for example, 3 s, 5 s, 10 s). Kling V3 supports up to 15 s.
  • Resolution dropdown — appears on models that expose resolution presets.
  • Sound effects toggle — appears on models that support AI-generated audio (for example, Kling V3). When enabled, the model generates ambient sound to match the video.
  • Settings icon — opens the side panel for advanced configuration.
  • Run button — run the node.

​
Credit cost

The node header shows the estimated credit cost for the current settings (for example, ~56 cr). For per-second-priced models like Kling V3, the displayed cost reflects the selected duration (for example, 5 s × rate).

​
Side panel inspector

Click the Settings icon to open the side panel. From there you can:
  • Upload start and end reference images with large tile pickers
  • Edit the prompt in a full-width text area
  • Adjust advanced parameters exposed by the model
  • Choose output format settings

​
Generation history

Past results are stored on the node. Use the arrow controls below the preview to navigate between generations, or click All for the full-screen gallery.
Chain an Image Generator into a Video Generator to animate a freshly generated image. Connect the image-out socket of the Image Generator to the image-in socket of the Video Generator.