Video Generation

Video generation AI creates moving video from a sentence or a single photo. OpenAI's Sora, Google's Veo, and Runway are the flagship names in this genre.

It's a far harder problem than image generation. Instead of one image, the model has to produce hundreds of frames, and the "physics of the world" — how a person walks, how water flows — has to stay consistent and natural across all of them. That means the model has to learn things like a dropped cup falling to the ground, and errors like stretched fingers or objects that randomly vanish still show up.

Even so, this is the fastest-moving corner of the field. Clips just a few seconds long are already being used in ads and music videos, and models have now reached the point of generating sound along with the picture. That's also why stories about production costs in advertising and film, and pushback from actors' and voice actors' unions, keep following this technology around.

"Generates an 8-second video with sound from a single line of text" — a typical line from articles announcing new video models, where the polish of the demo clip becomes the headline itself.