Reve opens the layout template that structures its images

The Reve API exposes its image structure as addressable regions. Code agents can now modify a single zone in 4K.

Reve is opening its API and, with it, the layout model that underlies its images. This is the first time this layer has been released outside of its in-house application.

The laboratory's bet, placed two years ago, rests on a single idea: representing an image not as a final rendering but as a set of structured and hierarchical regions, much like code. Each zone then becomes addressable, with its shape, color, and semantics modifiable independently of the rest. Opening this layer is explicitly aimed at agents: until now, an agent could describe an image and command it, without ever working inside it. Since a Reve image is code, it can now read its structure, analyze the relationships between its parts, and change only a single element. A skill file accompanies the API to plug a code agent directly into it.

The offering is split in two. On one side, classic endpoints for creating and editing, where a single call covers most integrations, with native 4K output (sixteen megapixels, without an upscaling pass) and style or subject consistency from one rendering to another, to maintain a brand identity across an entire campaign. On the other side, experimental endpoints that separate layout creation, editing, and rendering. The benefit is practical: you only regenerate the step that changed, replay the same layout across a series of variations, or propagate a correction through a batch of images without recalculating everything.

Finally, Reve claims that dividing the image into regions increases the reasoning context allocated to each, resulting in greater accuracy for small text and busy scenes.