Footage without a camera
Video generation AI produces moving footage from a text prompt or a single image. It has advanced rapidly as the frontier beyond image generation, and now produces footage that can be hard to distinguish from live action.
The main services
- Sora from OpenAI, the most widely discussed
- Veo from Google, high quality and integrated with Gemini
- Runway, Pika, and Luma, widely used by creators
- Integration into video editing tools, which continues to spread
They differ in clip length, resolution, whether audio is included, and pricing. Underneath, they use diffusion-family techniques much like image generation.
What it is used for
Short-form advertising and social video, product imagery, storyboards and previsualization, and shots that would be difficult to film. Being able to make footage without shooting it is changing assumptions about the cost and speed of video production.
Limits and cautions
Physically implausible motion still appears — wrong numbers of fingers, objects passing through each other — and long footage with consistent narrative remains a work in progress. Because convincing fake footage of real people is now possible, countermeasures and rules for labeling and watermarking AI-generated video are a live issue worldwide. Copyright questions are even hotter here than for images.
Length and quality improve every year, and complete short advertisements made entirely with AI already exist. Anyone who works with video, and increasingly anyone who watches video at all, needs the reflex of asking whether something might be generated.
Trying a short clip on a service’s free tier is the easiest way to get a feel for it.