Oracle introduces image and video AI models in a select-customer beta

Oracle is giving select customers tools to generate video, create images and edit existing pictures inside their own applications. Introduced at Oracle AI World on October 27, 2026, Oracle AI Media Models enter beta with a business-focused pitch: models Oracle says it trained on rights-cleared data, alongside services for customers’ authorized content.

Oracle calls this its first family of foundation models—the underlying AI systems used to build applications. The release brings together three distinct creative tasks, rather than presenting a single application that automatically takes a project from an idea through approval.

Making a scene versus changing one

• Video generation starts with a text description. Oracle’s example asks for a camera moving over a blue lake beneath snowy mountains, allowing a team to explore a motion concept for a retreat campaign.
• Image generation takes a written description of a subject, setting or visual style. A team could request the same mountain lake in cool daylight, then try a different prompt for warm evening light.
• Image editing starts with one existing picture and one written instruction. Oracle describes adding a canoe, removing an unwanted object or changing the canoe’s color.

Developers connect to the models through managed endpoints: access points where their applications send requests and receive results. Teams can connect the separate tasks in their own software. The organization decides how outputs pass through creative review, asset management and approval; those steps remain part of its workflow.

Training data and customer content

Oracle says it developed the models itself and trained them from scratch on rights-cleared data, using Oracle Cloud Infrastructure clusters optimized for multimodal and generative AI. That is the company’s account of the training process, not an independent assessment of the data’s permissions.

Alongside the models, Oracle offers data preparation that pays attention to media sources and usage permissions. It also describes model customization around customers’ content and creative requirements. Its cloud infrastructure supplies computing resources for curating media and training models at scale, with these services brought together in the OCI environment.

For generative AI to become a trusted tool in the entertainment industry, the quality and integrity of the data used to train models are critical

Daniel Mandell, Shutterstock SVP of Data Licensing and AI, in Oracle’s announcement

Finding footage instead of generating it

The announcement also covers media discovery: searching existing image and video libraries by meaning, including scenes, subjects and actions. Unlike generation, this is aimed at locating assets an organization already holds. Teams can use it to identify material that matches a creative brief.

Oracle illustrates the difference with a search for “a peaceful alpine escape.” Rather than requesting a new image, a team would use that phrase to find lake scenes, snowy peaks or canoeing footage in its archive for creative review.

Shutterstock, which Oracle identifies as a data partner, says it is exploring the generation API and media search and discovery capabilities for its collections. Its stated interests include creating, curating, organizing and searching multimodal content. That describes an exploration, not a confirmed production deployment.

Access remains through the select-customer beta. Oracle directs interested customers to their account teams to discuss evaluation access, supported capabilities and integration requirements. Its getting-started guidance distinguishes that process from signing up for a free OCI account: access to the media models is through the beta program.