Give motion room with input canvas

Use it before a subject hits the edge

Input canvas prepares the images sent to the video model. It scales the whole source image and adds space around it. This is useful when a character needs room to raise an arm, jump, or swing an object.

It does not generate new scenery, isolate an opaque subject, or reconstruct content already cropped from the original. Keep an uncropped source image whenever possible.

Supported modes and controls

Input canvas works with first-frame and first + last frame generation. The same configuration is applied to both endpoints. It is not supported for reference-image mode or video import.

Enable the input-canvas control in Studio, then choose:

Auto uses transparency for images with meaningful Alpha; otherwise it estimates a background from the image corners. A solid fill uses a six-digit RGB color. For Alpha Key generation, the extension stays transparent during preparation, before the temporary key background is applied.

Keep the two canvases distinct

Input canvas changes what the AI sees before generation. Output canvas changes how existing frames are placed during export. Changing the latter cannot recover a motion that was already generated outside the original picture.

The original upload is not overwritten and input padding is not a separate uploaded resource. The padding step does not add a separate generation charge, but generating the prepared request still follows the current quote.

Public API and MCP fields

This is a fragment of the shared animation request, not a complete runnable request:

{
  "video": {
    "input_canvas": {
      "enabled": true,
      "source_scale": 0.67,
      "position_x": 0.5,
      "position_y": 0.5,
      "aspect_ratio": "follow_output",
      "background": "auto"
    }
  }
}

source_scale accepts 0.5–1.0. position_x and position_y accept 0–1 within the available blank space: 0 is left/top, 0.5 is centered, and 1 is right/bottom. These are not pivot coordinates. The omitted object means no input-canvas configuration; use an explicit enabled value when storing presets.

Inspect both prepared poses, confirm model compatibility, and quote the final request. For export alignment and game-engine pivots, see canvas and pivot.