One technology, two different stories
The AI that makes a talking avatar for an online course and the AI that makes a fake video of a famous person are the same technology. The difference is not in the tool but in two things:
- Consent. Does the person know a digital copy of them is being created, and do they agree.
- Purpose. Is the content created to mislead someone.
An avatar made with consent for an internal video is a working tool. A video with someone else’s face, made without consent and passed off as real, is a deepfake.
What already applies
In Russia a citizen’s image is protected by civil law (Article 152.1 of the Civil Code): a person’s image may be made public and used only with their consent, with a small number of exceptions. This rule existed long before AI, and generation does not cancel it.
What this means in practice:
- You cannot make an avatar from an employee’s photo without their consent, even if you took the photo.
- You cannot use the face of a celebrity, blogger or actor in your content.
- You cannot animate someone else’s photo from the internet and publish it either.
In addition, if a deepfake is used for fraud, defamation or another offence, it is punished under existing articles, regardless of whether the law has a separate word «deepfake».
What is being discussed in 2026
At the time of writing there is no separate law on deepfakes, but there are several initiatives.
Senator Kozhin’s bill (March 2026) proposes banning the creation of deepfakes without the person’s consent and introducing serious liability, up to imprisonment and large fines. It also proposes a definition for the first time: a deepfake is false images, videos and voice recordings of citizens created without their consent. More about the bill.
Initiatives in the State Duma propose making the use of a deepfake an aggravating feature for a number of crimes (meaning a harsher penalty if a crime is committed with a deepfake) and separately protecting a person’s voice.
In spring 2026 a law was passed regulating the use of images and voices in campaign materials.
Important: bills are not yet law. But the direction is clear: regulation is tightening, and betting on «there is no separate law yet» is not the best strategy.
What the models themselves do
Regardless of the law, AI developers already put filters in place. Many models refuse to work with realistic photos of people, and all the more so with recognisable celebrities. For example, Seedance 2.0 is explicitly not designed for photos of real people.
Policies differ between models, so if one refuses a legitimate task (for example, an employee’s avatar with their consent), another may accept it. But getting around filters with rewording is not worth it: it does not work and risks getting the account blocked. More in the article on how to animate a photo with AI.
Rules for business
The practical minimum for working calmly with people in AI content:
- By default, generate people who do not exist. A synthetic character has no image rights, and model filters do not pick on it. For most tasks (an instructions presenter, a review host, a model wearing clothes) you do not need a real person.
- A real person only with written consent. The consent should describe where and how the image is used, for how long, and what happens if the employee leaves.
- No celebrities or public figures. Not in ads, not «as a joke».
- Do not mislead. If a viewer could mistake a generated person for a real one, especially in reviews or «documentary» videos, you are already in the risk zone. An honest label helps here; more in the article on AI content labelling.
- Keep your generation history and consents. If a question arises, you will have the evidence.
Separately on voice: AI voice cloning. On brands and characters: other brands, logos and characters in AI.
How to make your own synthetic presenter
If you need a permanent character, the most reliable route is to create one from scratch:
- Generate the face of a non-existent person in an image model.
- Build a set of angles in an editing model like Nano Banana Pro.
- Use that set as a reference in every video.
The process is described in detail in the articles on character consistency and on the AI talking avatar.
Why models refuse requests with faces: why AI refuses to generate.
In short
An avatar and a deepfake are one technology, and the line runs along consent and purpose. A person’s image is already protected by law: it cannot be used without consent. In 2026 bills with tough liability for deepfakes are being discussed, and regulation is clearly tightening. The calmest route for a business is synthetic characters, with real people only with written consent and no attempts to deceive anyone. This is not legal advice: in complex cases it is better to ask a lawyer.
