Reve is an AI image generation model and platform developed by Reve AI, Inc., a team of researchers, creatives, and engineers based in Palo Alto, California, founded in 2023 by Michael Gharbi, Taesung Park, and Mike Speiser. Its first model, Reve Image 1.0 (codenamed Halfmoon), launched in March 2025, and Reve 2.0 was released in June 2026.
Reve 2.0 takes a layout-first approach: it represents images as structured code describing composition, element positions, relationships, and style before rendering, separating planning from rendering. This design enables lossless iterative editing, direct manipulation of individual image elements, and agent-readable images. The model supports text-to-image generation, natural-language editing, and multi-image remixing with reference composition, with strong prompt adherence and in-image text and typography. It outputs native 4K images (4096x4096) without upscaling.
Reve ranked second on the Text-to-Image Arena leaderboard as of June 2026. The model is proprietary and available through the web app at reve.com and a beta API at api.reve.com with credit-based billing, plus hosting partners.
A free tier provides daily generation energy. Paid plans include Lite at $7.99/month and Pro at $19.99/month with expanded generation limits and storage.