Sora is an AI model that generates short, realistic video scenes from textual descriptions. Prompts can specify artistic styles, imaginary imagery, or real-world situations while guiding the scene’s content and movement. section 1.3.7
Sora uses a diffusion process rather than directly predicting each frame: it starts with noisy video frames and progressively removes noise to produce detailed, visually coherent output aligned with the prompt. section 5.4
Related video-generation approaches include:
- Image-to-video synthesis creates a video from a source image by matching learned image features to visual concepts. section 5.4
- Video-to-video synthesis transforms an existing video while preserving its underlying structure and changing its content. section 5.4
- Other named generators include Runway Gen-2, Pika, Hotshot, and Stable Video Diffusion. section 1.3.7
Prompted scene generation
A user can describe a realistic situation, an imaginary image, or a specific artistic style in a prompt, and Sora can generate a short video clip that follows those instructions while maintaining character and movement coherence. section 1.3.7
