Google researchers and early testers say the company’s Veo 3 video-generation model has begun displaying an emergent capability that lets users steer scenes by drawing directly on the first frame of a clip. By marking regions and overlaying short text instructions—an approach the community is calling “spatial prompting”—Veo 3 deletes the guidance frame and then carries out the directions in sequence across the generated footage. The technique gives creators finer, more intuitive control than traditional text-only prompts, which often require trial-and-error wording to achieve precise results. Demonstrations posted on 25 July show users replacing objects, inserting dialogue and restructuring scenes simply by labeling areas of the opening frame with commands such as “scene transforms instantly” or “remove text and execute in order.” Google Labs, which offers Veo through its experimental Flow interface, has begun circulating tip sheets that encourage annotating characters and scripts on frames to obtain accurate dialogue and staging. The company has not announced when or whether the feature will formalize into a product release, but the discovery underscores how generative-AI systems can develop unexpected ways for humans to direct complex visual tasks.