Sora
OpenAI’s Sora video model family — what the original introduced, how it differs from Sora 2, and where it fits today.
Sora is OpenAI's video generation family. The name draws heavy search interest on its own, and Higgsfield surfaces Sora models among its video options.
What Sora introduced
Sora's arrival moved the category's baseline expectation from "recognisably synthetic motion" to "plausibly filmed". The specific advances that mattered:
- Longer coherent shots than the models before it, where subjects survived the duration rather than dissolving.
- Physical plausibility — motion that respects weight and momentum instead of sliding.
- Consistent 3D space, so a scene held together as the camera moved through it.
- Natural-language prompting without rigid syntax.
Sora and Sora 2
Sora 2 is the newer generation and improves on the original across motion quality and prompt adherence. Both names remain in circulation, and different platforms expose different versions — worth confirming which one a given surface actually calls rather than assuming the latest. See Sora 2 for the current generation.
What has not changed
The category-wide limits apply to Sora as much as anything else: hands and fine manipulation remain unreliable, on-screen text warps between frames, quality degrades over longer durations, and no amount of description will reproduce a specific real person.
Content restrictions are enforced at the model. OpenAI applies limits around real people and public figures. These apply regardless of which platform you access Sora through, so a restriction you hit is not the platform being cautious.
Common questions
What is the difference between Sora and Sora 2?
Sora 2 is the newer generation with better motion quality and prompt adherence. Both names circulate, so confirm which version a platform actually calls.
Who makes Sora?
OpenAI.
Can Sora make videos of celebrities?
No. Restrictions around real people and public figures are enforced by OpenAI at the model level, not by the platform you use.