The Expand action lets AI build on an existing piece of content while still respecting its original form or intent. Use cases where this action is appropriate include extending the boundaries or ratio of an image, constructing a larger generation from a clip, or building upon a concept.
People don't always have an end goal in mind when in generative mode. Expanding takes a starting point and then builds upon it, often in multiple directions. Throughout this, the initial seed is kept intact, allowing creative exploration while maintaining connection with the original.
Expand starts with a starting point, which might include a previous generation, a rough draft, audio or video snippets, or a placeholder prompt. Users should be able to lengthen the original content while remaining true to the original form, or modify the new content through revised prompts, references, and parameters.
This supports an iterative process, and limits how much compute power is used on early concepts or verifying details. Users might use smaller or less refined generations as drafts and work towards a larger construction through multiple expand actions, paired with inpainting and other more convergent actions to fine-tune details as the work grows.
Expansion takes different shapes depending on the medium:
At its core, expand turns fragments into wholes. It is a copilot pattern that keeps users in control while exploring what comes next.
