Human review requested: Veo 3.1 reference images and allow_adult for a fictional 17-year-old character

Gemini API
1

Hello Google Gemini API team,

I am requesting a human-reviewed technical and product-policy clarification
for a non-sexual fictional-character workflow using the Gemini Developer API.

PLATFORM

  • Google AI Studio
  • Gemini Developer API
  • Model: veo-3.1-generate-preview
  • This is not currently a Vertex AI or Gemini Enterprise Agent Platform project.

USE CASE

We are preparing one isolated benchmark using up to three AI-generated
reference images of KAEL.

KAEL is:

  • a completely fictional and stylized webtoon character;
  • canonically 17 years old;
  • not based on, identifiable as, or similar to a real person;
  • not a public figure;
  • used in a non-sexual psychological-horror reaction shot;
  • not depicted with nudity, sexual content, exploitation, or graphic violence.

The Gemini API Veo documentation states:

  • personGeneration=“allow_all” for text-to-video;
  • personGeneration=“allow_adult” only for image-to-video, interpolation,
    and reference-image modes.

Gemini Cloud Assist indicated that the reference-image workflow would likely
be blocked because KAEL is canonically 17, but Cloud Assist also explicitly
confirmed that its response is not an authoritative Google product-policy
determination.

QUESTIONS

  1. Does personGeneration=“allow_adult” categorically prohibit the use of
    AI-generated reference images of a fully fictional and stylized
    17-year-old character in a non-sexual context?

  2. Is an allow_all project allowlist available for reference-image generation
    through the Gemini Developer API, rather than Vertex AI or the Gemini
    Enterprise Agent Platform?

  3. If no such allowlist exists for the Gemini Developer API, is there any
    compliant reference-image workflow that preserves the fictional
    character’s identity and canonical age?

  4. Which official Google channel can provide a human-reviewed determination
    for this specific Gemini API workflow?

COMPLIANCE POSITION

We are not seeking to bypass safety filters.

We will not:

  • misrepresent the character as 18 or older;
  • remove or conceal the canonical age to bypass a classifier;
  • alter the character to appear unambiguously adult;
  • upload the reference images;
  • submit a generation request;

until the permitted workflow is clarified.

No API key, project ID, payment information, or reference image is included
in this public post.

Thank you.

2

Hello @Les_Ciseaux_D_or_De ,

Does personGeneration="allow_adult" categorically prohibit the use of AI-generated reference images of a fully fictional and stylized 17-year-old character in a non-sexual context? → Just by looking at the reference image, the model won’t be able to consistently infer that the character is 17 years old (i.e. a minor). If the model recognizes the character as a minor, it will block the generation. However, it won’t block your account or API key.

Is an allow_all project allowlist available for reference-image generation through the Gemini Developer API, rather than Vertex AI or the Gemini Enterprise Agent Platform? → No

If no such allowlist exists for the Gemini Developer API, is there any compliant reference-image workflow that preserves the fictional character's identity and canonical age? → Usually the latest veo models will keep the character consistency by default, let us know if you observe otherwise. You can refer examples here

Which official Google channel can provide a human-reviewed determination for this specific Gemini API workflow? → No. As mentioned before, It is not possible to give deterministic advice on your workflow as it will vary on each generation.