Reve

Reve is a browser creative workspace built around the principle that creative intent should matter more than technical proficiency. Its in-house research team trains an ensemble of imaging models from the ground up using a novel architecture and a structured intermediate representation. Users can describe an image in a conversational interface, optionally attach reference images, choose an aspect ratio, and request one or more results. Reve can also edit generated or uploaded images through natural-language instructions, such as changing colors, objects, composition, or visual details, and it can remix multiple images into a new result. A drag-and-drop editor, identified by Reve as beta, provides direct manipulation of visual elements, including moving, resizing, recoloring, and rendering objects. The product also supports visual effects through presets or custom shaders. Video mode uses Google’s Veo integration: users can create clips from prompts or use selected images as references or exact start and end frames. Reve’s API exposes image generation, editing, and remixing models for partner workflows; its FAQ says video is not currently supported through the API. Created work can be downloaded or shared with a Reve link, with visibility and model-training choices varying by plan and account settings. The workspace therefore combines fast generative ideation with progressively more precise ways to control and refine visual output. The comparison workflow is especially valuable when a single prompt can be sent to several models and the decision depends on output quality rather than brand reputation. Users can use the available categories to narrow evaluation to the kind of artifact they actually need, compare alternatives visually, and return to leaderboards when a new model or benchmark result becomes relevant. This makes the site both a practical discovery surface and an ongoing reference for generative AI evaluation. The comparison workflow is especially valuable when a single prompt can be sent to several models and the decision depends on output quality rather than brand reputation. Users can use the available categories to narrow evaluation to the kind of artifact they actually need, compare alternatives visually, and return to leaderboards when a new model or benchmark result becomes relevant. This makes the site both a practical discovery surface and an ongoing reference for generative AI evaluation.