Reve

Reve 2.1

Reve 2.1 is a controllable text-to-image generation and editing model from Reve AI, an independent foundation-model lab. It is built on the idea that images should be composed like code — as hierarchical, structured regions — so that layout and control are core to the model rather than an afterthought.

Highlights

• Native 4K output. Renders dense scenes, fine text, and intricate structure directly at 4K (16 megapixels), preserving detail through iterative refinement.
• Layout planning. Reasons about structure, hierarchy, and spatial relationships before rendering, so it holds up on complicated, densely populated scenes.
• Editability. Every element is addressable — a single region can be changed and re-rendered without redoing the whole image.
• Multilingual text. Renders legible typography, including foreign scripts embedded directly in the image.

Reve 2.1 ranks as a top 4K model on the Arena and Design Arena leaderboards while using a fraction of the compute of comparably ranked models.

Model and imagery are the property of Reve AI, Inc. Learn more in the Reve 2.1 launch announcement. Use is subject to the Reve AI Terms of Service.