Sorry, we can't create videos from input images containing celebrity or their likenesses. — Veo API

Published 2026-09-23

raiMediaFilteredReasons Do not retry as-is

The exact error

Sorry, we can't create videos from input images containing celebrity or their likenesses. Please remove the reference and try again.

You asked Veo to animate a photo, and it decided the person in that photo is, or looks like, a well-known person. The job produces no video. Most people who hit this are working with their own product or model photos, not pictures of famous people.

What it means

Veo screens input images as well as outputs. Google’s Responsible AI page for Veo on Vertex (now Gemini Enterprise Agent Platform) has a support-code table with a Celebrity category, described as rejecting “requests to generate a photorealistic representation of a prominent person or if the project isn’t on the allowlist for this feature.” Two things follow from that wording:

The check is about a prominent person, and the message says “or their likenesses”, so an image can trip it because a face resembles someone, not only because it is that person.
Some projects are on an allowlist for this feature. Google’s page does not describe how a project gets on it.

On the Gemini API this message comes back in rai_media_filtered_reasons after the operation completes, as in forum thread 107802. There, a user generating videos from their own e-commerce photos with Veo 3.1 hit it repeatedly, said no celebrities were involved, and asked how to get on the allowlist; another user reported the same on some of their pictures. In thread 109707, a user of veo-3.1-generate-preview got it for a clothing-model photo of a non-celebrity. A community member (not Google) replied that Veo is very cautious with real-looking people and that a professional model may still be treated as recognizable. That is an opinion, not a documented rule.

How to confirm it’s this

The request included an input image (image-to-video, first and last frames, or reference images).
The operation is done, no video is returned, and the filter reason is this sentence.
On Vertex, the Responsible AI page maps support codes 29310472 and 15236754 to the Celebrity category. If your Vertex error carries one of those codes, it is the same filter. Google’s page does not say that the Gemini API message maps to a specific code, so do not expect one there.

Fix

Check the image, not the prompt. The message names the input image. Rewording the prompt is not what it asks for; it asks you to “remove the reference”.
Make sure you have rights to the image. If the photo shows a real, identifiable person, confirm you have their consent and the right to use the image. The community reply in thread 109707 raised copyright as a separate concern for photos taken from other websites.
Use an image you own of a person who agreed to it. If the filter still flags a photo of an ordinary, consenting person, that is a false positive worth reporting. Google’s filter messages ask for feedback when you think the result is wrong. Keep the image and request details for that report.
Ask about the allowlist through official channels. Google’s page says the Celebrity rule applies unless the project is on the allowlist, but it does not publish an application process. The users asking in the forum threads did not get an answer about access there. For a related Vertex case, a forum responder in thread 174130 pointed users to their Google Cloud sales representative or the Google Cloud forum.
Check personGeneration. Google’s Veo docs say image-to-video, interpolation and reference images accept only "allow_adult" on Veo 3.1, and that in EU, UK, CH and MENA locations allow_adult is the only allowed value. That setting controls people generally; it does not switch off the Celebrity check.

This page does not cover altering faces or images to get them past the filter. That would be working around a safety control.

Retry or not

No. The decision is about the image you sent. The forum reports describe the same photos being rejected repeatedly, and nothing in Google’s wording suggests a transient cause. Retrying the identical request mostly spends quota.

For agent code, stop on this reason and return a clear status saying the input image was rejected as a likeness. If your agent picks images automatically, send a different image rather than the same one again.

When it’s not this

This prompt might violate our policies about generating prominent people. Reported on Vertex in forum thread 174130. It points at the prompt rather than the input image, and the reporter says it appeared suddenly for assets that had worked before.
A support code in the Child range (58061214, 17301594). That category is about content depicting children and the personGeneration setting, not likeness.
videos were filtered out because they violated Google's Responsible AI practices. A block on the generated output, not the input image.
The audio message. Veo 3.x can drop a clip because of its generated audio, which has nothing to do with faces.

Error text last verified against the sources below on 2026-09-23.