Reve launches adaptive templates that rebuild layouts around brand assets

Why it matters

Templates gives Reve a route from benchmark-performing image models into recurring commercial design work, where control, brand consistency, and repeatable production determine whether teams keep paying.

Michael Gharbi and Taesung Park, two former Adobe Research scientists who co-founded Reve, launched an adaptive template system on Monday that lets users replace a product, logo, image, or copy while the surrounding design automatically adjusts.

Reve announced Templates in a two-post thread on X and published a more detailed product description on July 20th. Users can start with a premade design or create their own, choose an aspect ratio, and replace its core elements without rebuilding the composition from scratch. Reve says changes to the subject or purpose flow through the rest of the image, including its layout, color, type, and visual style. (x.com)

The launch turns the founders' technical bet on structured image generation into a workflow aimed squarely at advertising, e-commerce, and brand design. Templates currently cover categories including fashion, beauty, jewelry, food and drink, homeware, and automotive. Reve's example is deliberately commercial: a user can exchange a soda can for a perfume bottle, then let the system harmonize the surrounding design around the new product.

The model edits a structure, rather than repainting blindly

Gharbi and Park built Reve around a different internal representation for generated images. Instead of passing a text description directly into an image renderer, Reve first creates a structured layout containing regions with their own positions, sizes, descriptions, colors, and references. Users can then modify individual regions before rendering the result.

Reve detailed that architecture in The Layout Bet, published alongside Reve 2.0 on June 3rd. The Palo Alto company describes its layout as a code-like intermediary between a prompt and the final pixels. That structure is what allows Templates to preserve the basic design while replacing the product, headline, occasion, or imagery inside it. (blog.reve.com)

Gharbi's background is in computational photography, computer graphics, and computer vision. He completed a doctorate at MIT's Computer Science and Artificial Intelligence Laboratory before joining Adobe Research. Park studied mathematics and computer science at Stanford, earned his computer science doctorate at the University of California, Berkeley, and worked on generative image editing at Adobe Research. His prior work included the upscaling pipeline for Adobe Firefly and research behind image-generation and editing systems including GauGAN.

Their product strategy carries a pointed competitive implication. Two former Adobe researchers are packaging their own image model, editor, collaboration layer, and templates into one system. Adobe Express already lets customers generate editable templates from text, insert their own images, and customize branding. Reve is approaching the same market from its underlying model architecture: the template is represented as editable visual structure before the final image is rendered. Adobe approaches templates from a large existing design suite and content library. (helpx.adobe.com)

Reve is shipping the commercial layer around its model

Templates caps a concentrated run of releases. Reve launched version 2.1 of its image model on July 9th, opened the model and its layout controls through an API on July 14th, and introduced shared team workspaces on July 17th. The company also added Live Layers earlier in July, allowing users to place adjustable text, logos, shapes, and graphics over an image. (blog.reve.com)

That sequence shows where Reve is trying to build a business. The model supplies image quality, while Templates, Teams, Live Layers, references, and API access make the output usable inside recurring production work. A strong image generator can win benchmark attention. A shared template that preserves product accuracy and brand direction can become part of a marketing team's daily process.

Reve's launch materials leave several commercial details unspecified. The company has not disclosed how many templates are available, which subscription tiers include them, or whether teams can lock template elements and brand rules for less technical users. Those controls matter in the enterprise market, where templates often exist to prevent unauthorized changes as much as to accelerate creation.

Templates are available through Reve's web application. The feature gives Gharbi and Park a practical test of the architecture they have spent two years building: whether structured image generation can produce repeatable design systems, rather than another stream of isolated AI pictures.