Video Generate

Video Generate is the main generative video node. It combines a model, a prompt, the media the model works from, and the output settings in one place. Its clip lands in a Video Preview node and can continue into another video tool, such as Video Upscale or Video Trim.

The controls change with the selected model. A model that does not make sound, take a start frame, or edit an existing clip does not show those options. Video models go one step further than image models: their settings constrain each other, so a combination the model cannot produce is never offered, and the node says why.

1. 1

   Add Video Generate

   Find Video Generate in the Generative group and add it to the canvas.
2. 2

   Choose a model

   Open the model menu. Models are grouped by provider and show what they take: references, a start frame, or a clip to edit.
3. 3

   Pick the input shape

   Use the tabs above the media area: References, Start Frame, Edit Video or Extend Video. A plain text prompt needs nothing in the References tab.
4. 4

   Write the prompt

   Describe the subject, the motion, the camera and the mood. Name your references by number when the model reads them that way.
5. 5

   Set the output

   Choose the ratio, resolution and duration the model offers, and whether it should generate sound.
6. 6

   Press Run

   Review the credit estimate on the button, then run the node. The editor creates a missing Video Preview automatically.

A video model takes its media in one of three shapes, and never a mix. The node shows one shape at a time as tabs, and only the tabs the selected model supports.

References

A pool of pictures, clips or audio tracks the prompt refers to by number, such as “the product in image 1”. Leave it empty for a plain text prompt.

Start Frame

One picture the clip opens on. Models that accept an end frame show a second slot, and the clip travels from one to the other.

Edit Video

One clip to change. The prompt says what changes; references beside it supply a new look, object or style where the model allows.

Extend Video

One clip to continue. The new footage starts from the clip's last frame, and the sound follows the attached video.

Model

Chooses the video model and determines every tab and control shown below it.

Prompt

Describes the clip. A connected text input makes this field read-only because the upstream node owns the text.

Shots

On models that cut between shots, switches the prompt into rows, one per shot. The node compiles the rows into the model's own grammar at run time.

Ratio

Sets the frame shape from the list the model offers. In Edit and Extend the ratio can be fixed by the attached clip.

Resolution

Appears for models with more than one size. Some sizes rule out other options, such as sound or an end frame, and the node says so.

Duration

Chooses the length from the model's list. Some models offer Auto, which lets the model decide the length from the prompt.

Sound

Turns generated sound on or off where the model can make it. When editing or extending a clip, a third choice keeps the original audio.

Enhance

Improves a non-empty local prompt for the selected model. It is unavailable when a wire owns the prompt or Shots is on.

Run mode

Chooses whether a run adds a result or overwrites the wired preview.

Unlike image models, a video model's options depend on one another. One model generates sound only at its highest resolution and accepts an end frame only at a lower one, so the two cannot be combined. Another produces fixed lengths when it starts from a single frame.

The node resolves these rules before it shows the menus. An option that the current combination rules out is shown greyed with the reason, and the credit estimate on the Run button always reflects what will actually be sent.

Write a name inside square brackets, such as [Product name]. The node creates a text input port for that variable. Connect a Text Input or Text Generate node to supply its value at run time.

Each unique variable gets one port. Use it when the shot structure should stay fixed while a product, a line of copy or a location changes from run to run.

A normal run fills an empty wired preview first. If there is none, the editor creates one and wires it. Run Overwrite replaces the completed result in the wired preview, which keeps later connections intact.

On models that return the last frame, the node also offers it as a still on the Last frame port. Wire it into the next Video Generate's Start Frame to chain shots that flow into each other.

• Image Preview → Attachments

  Animate a finished image: it becomes the start frame, or a reference the prompt can name.

• Image Input → Attachments

  Keep a product, person or style in the clip from an uploaded picture.

• Video Input → Attachments

  Edit or extend a clip you already have.

• Video Generate → Video Preview

  Choose the exact holder that receives the clip.

• Last frame → Video Generate

  Start the next shot on the frame this one ended on.

• Video Generate → Video Upscale or Video Trim

  Finish the clip at a higher resolution or cut it to length.

• Video Generate trigger → another generator

  Start the next paid step only after this one succeeds.