openai / sora-2
sora-2
OpenAI's Sora 2 model for text-to-video or image-to-video generation.
Sora 2 represents a major leap in generative video and audio modeling, advancing toward realistic world simulation. The system builds on large-scale video pre- and post-training to achieve a deeper understanding of physical dynamics such as motion, rigidity, and momentum. Unlike prior models that distorted reality to satisfy prompts, Sora 2 maintains object permanence, models natural outcomes like rebounds and falls, and preserves coherent world states across complex, multi-shot instructions in cinematic, realistic, and anime styles.
Beyond visual fidelity, Sora 2 unifies multimodal control with synchronized speech, ambient audio, and sound effects. The model can also incorporate real-world footage—accurately reproducing the appearance and voice of any person, animal, or object from a brief reference video—allowing seamless integration of live and generated content.
Developers can now access Sora 2 through the Wiro platform API, enabling advanced experimentation with video, sound, and simulation generation. By integrating Sora 2 into their creative workflows, developers and researchers can prototype dynamic, physically grounded experiences that merge imagination with real-world consistency—pushing the boundaries of generative media and interactive storytelling.
Example prompts
Great starting points for sora-2.
API quick start
Run sora-2 with a single API call.
{
"prompt": "She turns around and smiles, then slowly …",
"seconds": 4,
"inputImage": "https://your-cdn.com/input.png",
"resolution": "720p"
}