Reve halts its creative tools following OpenAI's arrival

Reve will discontinue image and video generation on September 27, 2026. A portion of its research team is joining OpenAI, while the company prepares other AI-based products.

Ten days to finish creating, followed by just over a month to retrieve everything. Reve has announced that it is shutting down its image and video generation tools even as the company says it is accelerating its development with an investment from OpenAI.

Users will no longer be able to create new content on `reve.com` after September 27, 2026, at 11:59 p.m. Pacific Time. They will then have until October 31 at 11:59 p.m. to log in and export the images and videos stored in their libraries.

Reve has already stopped accepting new accounts and paid subscriptions. Registration closed on September 17 at 9:00 a.m. Pacific Time, before some users received the announcement.

Active subscribers will receive a full refund for their September payment. The company plans to return the funds to the original payment method by mid-October. Customers do not need to cancel their plans themselves to receive the refund.

The shutdown marks the end of Reve’s creative tools business, but not the company itself. Its product studio is moving away from visual generation and editing to work on other types of AI-powered workflows. Reve has not yet provided any examples of these future services or a timeline for introducing them.

That distinction matters. No full acquisition of Reve by OpenAI has been announced. The company says it will continue to operate independently, with its own product team and the freedom to work with different model providers.

Its relationship with OpenAI is nevertheless close. Reve announced on July 27 that it had received a new investment from the company led by Sam Altman. Neither company disclosed the size of the investment, the stake involved, or Reve’s valuation.

As part of the agreement, several members of Reve’s research team joined OpenAI to contribute to its multimodal research. The initial announcement said they would help lead those efforts. The latest message confirms that the move has now taken place.

Until now, Reve had described itself as two organizations operating under one name: a lab developing proprietary image models and a product studio turning that research into creative tools.

The company justified this structure by arguing that it needed control over both the model and the interface. Building its own technical foundations was supposed to make it possible to design an editor closely aligned with how images were represented, generated, and modified.

This approach produced a platform that combined creation, conversational editing, visual references, direct element manipulation, and a creative assistant. Reve sought to avoid separating the initial generation from the corrections later performed in traditional software.

Its main technical choice was to represent an image through a hierarchical structure called a “layout.” This structure described the different regions, their arrangement, and their content. Users could select an element, move it, replace text, or modify part of the composition without requesting a complete reconstruction of the scene.

Reve 2.0, launched on June 3, 2026, extended this approach across an editor rebuilt around layouts. The model produced images at native 4K resolution and was intended to improve text legibility, reference fidelity, and complex compositions.

A new version followed barely five weeks later. Reve 2.1, released on July 9, strengthened scene planning, multilingual text rendering, and the ability to modify individual regions separately.

On July 14, Reve also opened an API built around the model. Developers could use straightforward creation and editing interfaces or access the generation, modification, and rendering stages of a layout separately.

The speed of the change in direction is therefore striking. The Reve 2.1 API launched only two months before the creative tools shutdown was announced. The OpenAI investment had been disclosed less than eight weeks earlier.

The message sent to users explicitly states that generation on `reve.com` will end. The Help Center says more broadly that creating new images and videos will no longer be available after September 27. However, no public document separately explains what will happen to the Reve 2.1 API, its credits, or the professional integrations that rely on it.

Developers therefore cannot assume that the service will remain available after the web application closes. They will need direct confirmation from Reve before treating the API as a long-term dependency.

The future of Reve’s proprietary models is equally unclear. The company does not say whether their weights, licenses, or intellectual property have been transferred, shared with OpenAI, or retained exclusively by Reve. The investment and the hiring of part of the research team do not answer that question.

The closure affects more than images. Reve added video to its application in October 2025, although that feature relied on Google’s Veo 3.1. Users could produce a sequence from text or an opening image, add sound, and extend certain results. Reve supplied the creative environment rather than the underlying video model.

To retrieve their work, users must sign in to the web version and select the “Prepare Download” option displayed in the shutdown banner. The service then prepares one or more ZIP archives when a library is too large to fit into a single file.

The archive can include creations from both the web and mobile applications. Users can also select a specific period and repeat the process to retrieve items added after an earlier export.

Files uploaded by users as source material or references are not included. Reve’s Help Center states that the export contains only images and videos generated through the platform. Anyone who no longer has the original reference files will need to recover them separately.

Download links remain valid for two days. If a link expires, users can prepare another archive, but Reve limits this process to one request every 24 hours. Anyone with a large library would be wise not to wait until the final hours before