Reve 2.0

Reve 2.0 is a layout-first AI image model from Reve AI, released on June 3, 2026. Instead of turning a single sentence into pixels, it builds the image as a structured layout where every element has its own position, size, and description, which you can edit before and after rendering. It handles text-to-image, image editing, multi-reference composition, in-image text, and output up to 4K.

Yes. Reve 2.0 and Reve Image 2 refer to the same model, the second generation of Reve AI's image model. Reve labels the family Reve Image 2.0; most people shorten it to Reve 2.0.

Most image models go straight from a prompt to pixels, so a small change can ripple across the whole picture. Reve 2.0 inserts an editable layout in between, similar to how HTML sits between intent and a rendered web page. You can move, resize, or rewrite a single element and re-render just that, which makes results predictable and editable rather than one-shot.

Reve 2.0 renders natively at 2048 by 2048 and can upscale to 4096 by 4096, keeping most fine detail through the upscale. It supports seven aspect ratios, from 16:9 and 9:16 to 1:1, and exports as PNG, JPEG, or WebP.

Yes. Reve 2.0 runs a dedicated typography step so headlines, packaging copy, and labels come out legible, correctly spelled, and placed in the right perspective. That makes it well suited to posters, packaging, and UI mockups where text accuracy matters.

On launch day Reve 2.0 debuted second on the public Text-to-Image Arena leaderboard, just behind GPT Image 2. The practical difference is control: GPT Image 2 leads on raw prompt-to-image quality and reasoning, while Reve 2.0's edge is the editable layout, so you place and adjust elements directly rather than re-describing the whole scene.

Reve 2.0 is built by Reve AI, an independent image-AI lab based in Palo Alto. The model is available through Reve's own app and API and through hosting partners.