What AI Avatar Onboarding Data Do You Need?
Start with the use, tool, consent, minimum useful references, constraints, data controls, and a reviewed test rather than a universal photo count.
No single photo count or prompt fits every AI avatar. The right input depends on the tool, the use, the person, the output, consent, and risk. Read the provider's current rules first. Then collect only what the approved use needs.
This is a planning guide, not legal advice. Tools do not all work the same way. Their rules can change too. Before you upload personal data, check the contract and privacy terms. Also check identity rights, access needs, storage, deletion, and local law.
Start With the Intended Use and the Tool
Name the output, audience, channel, and language. Then note the region, life span, and review owner.
Read the tool's rules on input, quality, age, and consent. Then check business use and storage. Check training, export, and deletion too.
Ask if you need an avatar at all. A real video, voice clip, drawing, screen capture, or text may do the job. Those often use less personal data and carry less risk.
Confirm Rights, Consent, and Limits
List each face, voice, name, work, logo, place, and background asset in the project. Make sure the person knows each planned use and can approve it. A release is not endless permission. It does not cover every future channel, language, edit, or claim.
Write the limits before you collect files. You may bar health, legal, money, political, adult, false, unfair, or endorsement uses. Name who may ask for new work and who reviews it. Say how consent can change, and what happens when a person leaves the firm.
Collect the Smallest Useful Set
Follow the tool's rules for file type, frame, light, sound, length, and range. Use clear, current files that you have the right to use. More is not always better, and less may not be enough. Test the smallest set that meets the rules. Add data only when a written gap calls for it.
Identity: the approved face, voice, name, and how to say it. Add only the traits the use needs.
Style: clothing, setting, tone, and camera look. Add the brand rules. Mark examples as guides, not promises.
Release: audience, channel, language, shape, and length. Then add captions, a transcript, alt text, and review needs.
Exclusions: words, claims, goods, scenes, gestures, and backgrounds that must not appear. The same goes for face changes and sensitive topics.
Protect the Data From Start to End
Treat face and voice files as sensitive personal data. Limit access and use approved accounts. Record where source files, new files, prompts, and approvals live. Ask if the provider trains on uploads or shares them with vendors. Check that it can delete them and give useful export or audit records.
Set a delete date and an owner. Plan for a lost account, a vendor change, or a disputed result. Plan for withdrawn consent, a security event, or a removal request. The NIST AI risk guide supports risk checks across design, use, review, and change. Intake is not a one-time form.
Test Before a Wide Release
Run a small test that you can undo. Check the face, voice, meaning, and how names are said. Check the claims, disclosure, captions, transcript, and visual access. Check local context and brand fit too. Include the person shown, and skilled subject reviewers when needed. Record the input version, tool, settings, reviewer, choice, and fix.
A sharp image does not prove the result is true or safe. Generated media may add, drop, or change details. Ask a person to approve each key use. Take extra care with endorsements and regulated messages. The same goes for client contact. Take care too with claims that may affect someone.
Keep One Intake Record
Record the purpose, the approved channels, and the audience. Then list the regions, the owner, and the reviewers.
Tool, version, provider rules, input list, and rights basis.
Consent scope, banned uses, and disclosure. Also the plan for captions, transcript, and alt text.
Storage, access, and the training setting. Note the keep and delete dates. Note how consent can change, and the incident route.
Test cases, flaws, fixes, and final approval. Then the release date and the next review date.
How TTGC and Kyndrify Can Help
TTGC and Kyndrify sell avatar production services. Within an agreed scope, the team can help plan intake and prepare approved files. They can set brand and use limits, make test outputs, and record the review. Fit, controls, and results depend on the service and current terms. They also depend on source files, approvals, and the planned use.
Need a controlled avatar intake?
TTGC can help define the use, inputs, rights, review, access needs, and release record for an agreed avatar project. Results and legal fit are not promised.
Sources
- NIST — Artificial Intelligence Risk Management Framework. https://www.nist.gov/itl/ai-risk-management-framework
- Federal Trade Commission — Policy Statement on Biometric Information and Section 5 of the FTC Act. https://www.ftc.gov/legal-library/browse/policy-statement-federal-trade-commission-biometric-information-section-5-federal-trade-commission-act
- W3C Web Accessibility Initiative — Images Tutorial. https://www.w3.org/WAI/tutorials/images/
