Reve relaunches its image tool with an editor
Reve has relaunched Reve Image with natural-language editing, a drag-and-drop object editor, a creative assistant and an API, all built on what it calls a code-like "layout representation" of images.
Image-generation startup Reve has announced a rebuilt version of its Reve Image product. The company describes it as four products in a single interface:
- Creator and remixer: users can edit generated or uploaded images with natural-language instructions and combine elements from several images into one. Reve says its models are "best-in-class" at aesthetics, text rendering and detail, but the post gives no benchmarks.
- Drag-and-drop editor (beta): an object-based editor where users click, drag and re-render elements of an image.
- Creative assistant: a conversational collaborator that can search the web for inspiration and turn ideas into images.
- API (beta): third parties can use Reve's creation, editing and remixing models in their own applications.
The technical claim at the center of the launch is a new architecture. It breaks images down into what Reve calls a "layout representation," a structured, code-like intermediate format. According to the company, this format records spatial relationships and fine detail that a text prompt can't capture. It also lets edits keep a person's likeness, and it supports conversational editing alongside references from several images. Reve says it exposes this representation to users, which is what lets them move, resize, add, remove or replace individual objects. The company contrasts this with tools that act as black boxes.
The post does not include pricing for the API, model specifications or evaluation results. It does not say how the layout representation is generated or trained. Reve says the product is free to try at reve.com. The post is signed by co-founders Christian Cantrell, Michaël Gharbi, Taesung Park, Jon Watte and Mike Speiser.