Reve is an AI image generator and editor that lets you create, edit, and remix images by combining natural-language edits with a drag-and-drop editor. Where most image models take a prompt and hand back a flat picture, Reve is built on what the company calls a layout-first architecture: the model reasons about structure, hierarchy, and spatial relationships before it renders, organising the image into labelled, addressable regions. The practical consequence matters more than the theory. Because every element in the picture is separately addressable, you can change one object, one line of text, or one part of the background and re-render just that region — instead of rolling the dice on a whole new generation and hoping the parts you liked survive.
The second defining trait is resolution. Reve renders at native 4K x 4K resolution, a true 16 megapixels, with no upscaling step, which is why the product markets itself around production work rather than social thumbnails: graphic design, art direction, product photography, portraits, interior design, fashion, filmmaking, architecture, and post-production all appear on its own list of target scenarios. The current model generation is Reve 2.1, released in July 2026 just a month after Reve 2.0, and the company describes itself as an independent lab that trains at a fraction of the compute used by the larger labs it ranks alongside. Its earlier model, Reve Image 1.0 (codenamed Halfmoon), first drew attention when technology press reported it topping the Artificial Analysis Image Arena for image generation quality ahead of Midjourney, Imagen 3, Recraft and FLUX.
One piece of company news is worth knowing before you build a workflow on it: in July 2026 Reve announced an investment from OpenAI, with members of its AI research team joining OpenAI to lead multimodal research, while stating that Reve continues to operate independently and that existing products, including Reve.com, will continue.
Reve runs on three plans and a consumable allowance the company calls energy. The Free plan costs nothing: new accounts receive an introductory allocation of creative energy and a daily refresh thereafter, which makes it genuinely usable for evaluation rather than a one-shot trial. Paid tiers are Lite at 7.99 US dollars per month and Pro at 19.99 US dollars per month, plus tax where applicable, both billed monthly and cancellable at any time from the Account tab. Lite provides five times the energy of the free tier and Pro one hundred times, with storage scaling on the same multiples.
Video is metered separately from image energy. Free and Lite accounts receive only a one-time video allowance granted at sign-up, while Pro includes 250 video energy per month and additionally permits standard energy to be spent on video, capped at 100 per day. Two details deserve attention before you commit. First, energy is defined in the Terms of Service as a daily usage limit that depletes with use; Energy has no cash value and cannot be reloaded, resold, transferred, or redeemed, and unused amounts expire without a refund. Second, model-training treatment differs by tier: the plan comparison marks the free tier as default opt-in only, while Lite and Pro accounts get an opt-out. API usage is priced separately through the API console. Prices and allowances change as the product iterates, so confirm on the official plans page before subscribing.
Yes. The Free plan gives new accounts an introductory allocation of creative energy plus a daily refresh, so you can keep testing over time rather than burning a single trial. Paid plans are Lite at 7.99 dollars and Pro at 19.99 dollars per month, plus tax.
Yes. Reve states that you own the content you create and are free to use it for any lawful purpose, including commercially, on all three plans, and that you keep ownership of past work even if you unsubscribe. You remain responsible for ensuring the output does not infringe third-party rights.
Native 4K — 4096 by 4096, a true 16 megapixels rendered directly rather than upscaled from a smaller generation. That resolution holds through iterative edits.
The model plans a structured layout of labelled regions before rendering. Because each region is addressable, you can edit one element and re-render only that part, which gives far more predictable control over composition and text placement than prompt-and-pray regeneration.
Text rendering is one of the model's explicit design goals. Reve 2.1 produces dense, legible type inside the image and extends that to foreign scripts, which suits signage, packaging and editorial layouts. Proofread anyway before shipping.
Yes. The Reve 2.1 API provides ready-to-use create and edit endpoints plus experimental Full Layout Control endpoints that decouple layout creation, editing and rendering. A published skill.md lets coding agents work with images structurally. API pricing is separate from the subscription plans.
Yes, through an integration with Veo rather than an in-house video model. Veo 3.1 is the default with a Fast variant available, clips run 4, 6, or 8 seconds at 720p or 1080p, square format is unsupported, and audio is generated rather than uploaded.
It depends on your plan. Paid accounts can opt out of model training from the Account page, effective for content created after you enable it. The plan comparison lists the free tier as default opt-in only, so free users do not get that switch.
Not automatically on the free tier. Output created on a Free Account may be made public on and searchable through Reve's Inspiration page, with a perpetual licence granted to Reve and other users. Use a paid account for confidential work.
Energy is a daily usage limit that depletes as you generate and edit. Lite carries five times the free allowance and Pro one hundred times, with video metered separately. Energy has no cash value, cannot be transferred or resold, and expires unused without a refund.