OpenAI upgrades Video API with Sora 2 custom characters and 20s clips

Anyone who has tried building with AI video knows the pain: characters morph into new people every three seconds, and clips end before anything actually happens.

OpenAI is trying to fix that for developers. The company rolled out a fresh batch of upgrades to its Video API, all powered by the Sora 2 model.

Longer scenes and persistent characters

Developers can now generate video clips up to 20 seconds long directly through the API.

If that is still too short, a new video continuation feature lets you extend an existing scene instead of awkwardly splicing cuts together.

Even better, the API now supports custom characters and objects. That means you can finally keep the same actor or product consistent across multiple shots without playing prompt roulette every time.

Rounding out the release, OpenAI added export options for both 16:9 horizontal and 9:16 vertical aspect ratios, alongside batch jobs for spinning up generations at scale.

Why it matters

For developers building video apps or creators aiming for automated production, this is the shift from a tech demo to an actual workflow tool. Consistent characters and scene continuation are table stakes for any serious storytelling or ad creation.

That said, we think the real friction will still be reliability and cost. Batch generation is great, but if extended 20-second clips still hallucinate, developers could burn through credits fast. Still, OpenAI looks determined to turn Sora 2 into the default backend for the next wave of generative video products.