Oracle's first family of foundation models for media — image generation, image editing, and video generation delivered as API building blocks on OCI, pitched on rights-cleared training data.
Last verified · 2026-10-28 · by Moe Ameen
Oracle AI Media Models are Oracle's first family of foundation models for media, announced on October 27, 2026 at Oracle AI World and available in preview on Oracle Cloud Infrastructure (OCI). Unlike a consumer app you log into, they are building blocks: developers call them through managed endpoints and embed the capabilities in their own software, while the organization keeps control of creative review and approvals. Access is limited — Oracle frames it as a preview, and other coverage describes a select-customer beta arranged through Oracle account teams.
The family spans three generation capabilities plus search. Oracle Image Generation creates an image from a written description of a subject, setting, or style; Oracle Image Editing takes an existing image and a text instruction to add, remove, or modify elements; and Oracle Video Generation creates video from a text prompt. A media search capability lets teams find existing footage in a library by meaning rather than generating new content.
Oracle's headline differentiator is provenance: it says the models were trained from scratch on rights-cleared data, on OCI clusters built for multimodal and generative AI, and positions that as an IP-safety advantage for enterprises and studios. That training-data claim has not been independently verified. Shutterstock is named as a data partner exploring the generation API and media search, with no production deployment announced yet. The models are part of a broader platform, Oracle Media Fabric, which Oracle says is planned within the next 12 months and includes a model-agnostic Oracle Creative Studio.
The honest boundary: this is enterprise infrastructure, not a finished creator tool. Oracle disclosed no pricing and no technical specs — resolution, clip length, model size — at launch, and the models generate an asset and stop. Captioning, reframing, brand voice, scheduling, and publishing are all a separate job.
Oracle AI Media Models are designed for the people building the app, not the person posting to Instagram — you call an endpoint, get an asset back, and the rest of the content workflow is yours. [Kompozy](/) is the layer that takes a raw generated asset and turns it into finished, published content, and it does not care which model produced the asset. So the clean pattern for a developer or an agency with OCI access is: generate the image, edit, or clip through Oracle's endpoint, hand the file to Kompozy, and let it do everything between "asset exists" and "the audience saw it." That means auto-reframing one clip to 9:16, 1:1, and 16:9 with burned-in [captions](/glossary/caption), fanning a single image into a brand-exact [Carousel](/glossary/hyperframes) and [Quote Graphics](/glossary/output-buckets), wrapping the footage into a [Clipped Short](/glossary/clipped-short), and writing the matching blog and newsletter — every piece held to one voice by your [Persona Brief](/glossary/persona-brief).
The reason this split works is that the two tools solve non-overlapping problems. Oracle's models answer "can I generate a rights-cleared asset at scale on infrastructure I trust?" Kompozy answers "how do I turn any asset into a week of on-brand posts across every platform?" Kompozy also generates net-new formats Oracle's endpoints do not — [Persona Shorts](/glossary/persona-shorts) talking-head avatar video, [Persona Frames](/glossary/persona-frames), and VFX hooks — so the Oracle-generated scene becomes the B-roll inside a fuller piece. Then it schedules and publishes the whole set across the eight social platforms plus blog and email from a single review queue, on a gate or [Autopilot](/glossary/autopilot). Generate with Oracle; compose, brand, and ship with Kompozy.
They are Oracle's first family of foundation models for media, announced on October 27, 2026 and available in preview on OCI. They cover image generation, image editing, and video generation, plus semantic media search, delivered as API building blocks developers embed in their own applications rather than as a standalone app.
Not as a typical consumer product. The release is a limited preview on Oracle Cloud Infrastructure, with access arranged through Oracle account teams, and the capabilities ship as developer endpoints. They are aimed at enterprises, studios, and app builders. If you want self-serve generation plus publishing today, a tool like Kompozy is the more direct path.
Oracle's main claim is provenance: it says the models were trained from scratch on rights-cleared data and positions that as an IP-safety advantage for enterprises. It also offers a model-agnostic Oracle Creative Studio that can mix Oracle's models with third-party ones. Oracle did not disclose pricing or technical specifications at launch, and the training-data claim has not been independently verified.
Oracle's models generate an asset and stop — there is no captioning, reframing, or publishing. Pair them with a content engine like Kompozy: generate the image or clip via the OCI endpoint, then let Kompozy reframe it per platform, add captions, fan it into carousels and quote graphics, and schedule it across the eight social platforms plus blog and email.