Video Node
Choose a mode and preset before connecting images, video, or audio references.
Start with a Video Generation node
The Video menu offers Video Generation as well as separate Video Edit, Extract Frames, and eligible regeneration tools. Generating a new shot starts with Video Generation.
- Right-click blank canvas, choose Video → Video Generation, and left-click to place it.
- Select the Video node and write what moves over time, for example “camera tracks beside a red bicycle through light rain.”
Let the model determine its ports
Video’s input ports are model-dependent. Text, image, video, and audio are not universally interchangeable: a selected preset determines visible compatible inputs and reference limits.
- Open the Video mode and model controls before wiring media; inspect the labels on its left edge.
- For image-led motion, select a compatible model and connect the picture to an Image input; for text-led motion, supply a prompt or Text input.
Choose the references to send
Connected material can appear in Video’s @ picker after it has a usable output. Selection order is meaningful for numbered references such as @1 and @2 in the prompt.
- Connect a successful Reference image or uploaded Video/Audio node to a port that the current model supports.
- Open @ reference media, select only the relevant thumbnails, and match any @ numbers in the prompt to that order.
Check duration and ratio before running
Duration, aspect ratio, and resolution controls vary by preset. Some models derive the output ratio from a first-frame image or source video rather than a freely chosen ratio.
- Review the visible model-specific settings and its input notices; begin with a short supported duration.
- Run the Video node, wait for a usable preview, and inspect subject continuity before using the result downstream.
Fix incompatible combinations
For example, MiniMax H3 reference image mode requires one image and one audio clip, while its video mode uses videos with audio and does not mix those with separate image/audio inputs. These are preset-specific rules.
- If a run rejects inputs, re-read the selected model notice, inspect current port labels, and remove unsupported sources.
- After switching models, check whether prior connections or @ selections were remapped; run a single short test instead of the entire branch.
Model access, quota, and provider configuration vary by deployment; an option in the UI does not guarantee a successful render.