At Oracle AI World, Oracle introduced a family of foundation models for media — generating and editing images and generating video from text — available in preview on Oracle Cloud Infrastructure and pitched on rights-cleared training data.
2026-10-28 · by Moe Ameen
On October 27, 2026, at Oracle AI World in Las Vegas, Oracle introduced Oracle AI Media Models — what it describes as its first family of foundation models for media — available in preview on Oracle Cloud Infrastructure (OCI). The release is limited: Oracle's blog frames it as a preview, while other coverage describes a select-customer beta arranged through Oracle account teams rather than an open OCI sign-up.
The family covers three generation capabilities plus search. Oracle Video Generation creates video from a text prompt; Oracle Image Generation creates an image from a written description of a subject, setting, or visual style; and Oracle Image Editing takes an existing image plus a text instruction to add, remove, or modify elements. A media search capability rounds it out, letting teams find existing footage in their libraries by meaning rather than generating something new. Oracle ships these as building blocks: developers call them through managed endpoints and embed them in their own applications, and the organization keeps control of creative review and approvals.
Oracle says the models were trained from scratch on rights-cleared data, on OCI clusters built for multimodal and generative AI — a provenance and IP-safety pitch aimed at enterprises, studios, and brands. That training-data claim has not been independently assessed. Shutterstock is named as a data partner and is exploring the generation API and media search for its collections, though it has not announced a production deployment.
The models are one piece of a larger platform, Oracle Media Fabric, which Oracle says is planned to be available within the next 12 months. Within it, Oracle Creative Studio is a model-agnostic production suite that can mix Oracle's own models with leading third-party models, chosen per task by security, provenance, cost, and IP-protection requirements. Oracle disclosed no pricing and no technical specifications — resolution, clip length, or model size — for the media models at launch.
Oracle's media models sit behind an enterprise wall — an OCI account, a conversation with a sales team, and raw API endpoints you wire into your own app. Most creators will never touch them, and that gap is exactly what [Kompozy](/) already fills on the open web. It is a self-serve generation and publishing engine: point it at a source or an idea and it produces finished, on-brand content in 18 formats — [Persona Shorts](/glossary/persona-shorts) and avatar video, [carousels](/glossary/hyperframes), infographics, quote graphics, blogs, and newsletters — then schedules and publishes them across the eight social platforms plus blog and email from one queue on [Autopilot](/glossary/autopilot). No enterprise contract, no cloud console, no endpoint plumbing.
If Oracle's rights-cleared pitch tells you anything, it is that the valuable layer is no longer the raw generate-an-image call — that is becoming a commodity. The value is everything that turns an asset into a consistent, multi-platform presence: one [brand voice](/glossary/persona-brief) enforced across every piece, per-feed reframing to 9:16, 1:1, and 16:9 with burned-in captions, and a review gate before anything ships. When foundation models like Oracle's become callable building blocks, a tool like Kompozy is what assembles them into content a creator can actually post. For how the same enterprise shift is playing out elsewhere, see our coverage of [Adobe's creative agents](/news/adobe-creative-agent-expansion) and [Alibaba's Wan 3.0](/news/alibaba-wan-3-0-launch).
They are Oracle’s first family of foundation models for media, announced on October 27, 2026 and available in preview on Oracle Cloud Infrastructure. They cover three generation capabilities — image generation from text, image editing from an instruction, and video generation from a prompt — plus semantic media search over existing libraries, delivered as API building blocks developers embed in their own apps.
Not easily. The release is a limited preview on OCI, with access arranged through Oracle account teams rather than an open sign-up, and the models ship as developer endpoints rather than a finished creator app. They are aimed at enterprises, studios, and app builders. A self-serve tool like Kompozy is the practical path for an individual creator who wants to generate and publish content today.
It is the broader platform the media models belong to, which Oracle says is planned to be available within the next 12 months. It includes Oracle Creative Studio — a model-agnostic production suite that can mix Oracle’s models with third-party models — and Oracle Video Edge for streaming and delivery.
Oracle says the models were trained from scratch on rights-cleared data and positions that as an IP-safety advantage for enterprises. That claim has not been independently assessed. Shutterstock is a named data partner that is exploring the generation API and media search but has not announced a production deployment.