Real-person video is not just another image-to-video task. A face, voice, body, and context carry consent and likeness risk, especially as Gemini Omni-style workflows move toward avatars and conversational edits.
Use /image-to-video only when you have the right to use the source material and the review or publication context is clear.
Start with permission
Confirm who appears in the source, whether the image or voice may be used, where the result will be reviewed, and whether publishing is allowed. A generator can create a draft; it cannot grant rights.
Keep the first movement restrained
Start with one human-scale change: gentle push-in, natural blink, small head turn, slight shoulder shift, or subtle light movement. Avoid changing identity, voice, clothing, setting, and action in the same first pass.
Prompt pattern
Use the uploaded portrait only as the likeness anchor. Add one subtle movement: [movement]. Preserve face shape, hairstyle, outfit, and main silhouette. Do not imitate a private person, alter speech, or imply endorsement. Review: does the result remain respectful and authorized for the intended context?Review before sharing
Check consent, source rights, platform policy, disclosures, SynthID or provider labels, and whether viewers could mistake the result for real footage.
Takeaway
For real people, Gemini Omni is a review workflow first. Keep consent explicit, motion restrained, and publication criteria visible.
