A photo became the beginning of a generated clip
Google added a photo-to-video capability to Gemini in July 2025, letting eligible subscribers create eight-second clips with sound using Veo 3. Instead of describing everything from scratch, a user could upload an image and give instructions for the scene and audio.
The company announced a rollout for Google AI Pro and Ultra subscribers in selected countries. It also said generated videos would carry both a visible label and an invisible SynthID watermark. Those were provenance features in the announced workflow, not proof that every viewer would recognize a synthetic clip after it was shared elsewhere.
Flow received a related but separate update
On the same day, Google's Flow team described adding generated speech to its Frames to Video feature. That workflow used an uploaded image as a clip's opening frame. Sound effects and background noise had already been supported; speech was the new addition described in that announcement.
Flow also introduced Frames to Video for Veo 3 Fast. Google called audio generation experimental and warned that results could vary. Its availability update expanded Flow to more countries, while noting that individual features and supported image types could still differ. A product's geographic reach was therefore not a guarantee that every capability worked identically everywhere.
The starting image and the resulting footage are different evidence
The combined announcements show two interfaces applying the same broader idea: use a still image to guide a newly generated scene. The animation and accompanying sound are model output, not a recording of what happened around the original photograph. That distinction matters when a realistic-looking result travels without the prompt or source image that produced it.