Strategy & perspective / AI music
When AI video refuses a real person, treat it as a production signal
Why image-to-video models may refuse photoreal people, how consent and impersonation rules matter, and what alternatives musicians can use.
The short answer
If an AI video model refuses a photoreal person, do not treat it as a puzzle to evade. It is often a deepfake or consent guard. You can switch to a non-photoreal style, use a model and workflow that supports the shot with consent, film original material, or choose code video when the idea does not need a person.
A refusal is information
In Studio testing, Seedance refused image-to-video when the source frame showed a photoreal person. The app now explains that as a deepfake guard. That message can be frustrating, but it is also useful: the current shot asks the model to do something that raises identity and consent concerns.
Platform rules point the same way
YouTube requires disclosure when AI makes a real person appear to say or do something. Spotify says vocal impersonation should happen only with authorization from the impersonated artist. Those are different policies in different domains, but the creative principle is consistent: realistic identity changes require care, consent, and records.
Practical alternatives
Move away from photoreal likeness when the person is not essential. Use a non-photoreal character, sculpture, mask, silhouette, object, or environment. Film original footage when the person matters. Use a model and workflow that supports the shot with the right permissions. Or choose code video when the song needs motion, typography, and beat response more than a person.
Write the safer brief
A safer brief says what the shot communicates without relying on a real person's face. "A paper figure learns a repeated gesture from an ink shadow" may carry the same emotional role as a forbidden synthetic portrait. The goal is not to dodge policy. It is to find a visual language the release can own.
Connect the work
Go deeper before the next release decision.
Consent shapes the shot
Choose a visual route that does not fight the guardrail.
Use Studio to test non-photoreal, code-led, or consent-safe video directions while keeping provenance attached to every generated asset.
Plan a safer shotFrequently asked
Questions, answered.
- Why did my image-to-video job refuse a person?
- A photoreal person can trigger deepfake or consent protections, depending on the model and workflow.
- What should I do instead?
- Use consent-safe footage, a non-photoreal style, a model path that supports the use case, or code video when no person is needed.
Sources
