Image to Video now accepts an optional second reference image that anchors the final frame, so the clip transitions from the start image toward the end image over its duration.
It gives you control over transitions where text prompts fall short: a particular camera path, a specific final pose, or an exact scene composition.
A common fashion example: rotating a model to show the back of a garment. Text prompts fail because the video model can't infer the back from the front. With a back-view image as the end frame, the rotation renders the real garment instead of an invented one.
In the App
Available on Pro and higher plans in Studio. Setting an end frame automatically locks resolution to 1080p, the only resolution that supports this input.
For API Users
The image-to-video endpoint accepts a new optional end_image parameter (URL or base64). It must be paired with resolution: "1080p"; other resolutions are rejected with a validation error.