Using a Teacher's Voice and Image in AI Content: Permissions and Risks

A school has dozens of recorded lectures and permission to sell the course. The team wants to translate it into another language while retaining the teacher's voice, then create a short promotional video featuring their digital avatar. Technically, both tasks may seem like an extension of ordinary editing. For the person whose voice and likeness are used, this is a different situation: the system can create statements they have never made.

The question extends beyond whether the school owns the original video file. You need to distinguish rights in the materials, the interests of the person depicted, data processing, the AI provider's rules, and responsibility for the new video's content. A single clause in an old contract should not automatically be interpreted as unlimited permission for any future generation.

Below is a practical process for preparing such a project. It helps a school formulate a decision and give a lawyer specific terms rather than a general request to permit AI use. The final legal arrangement depends on the contracts, countries of distribution, data processing methods, and particular materials.

Distinguish Four Different Scenarios

Ordinary editing retains a real recording, although it may change its context. Synthetic dubbing creates a new audio track. Voice cloning allows new texts to be spoken in a similar voice. A digital avatar adds an image of the person apparently speaking those texts. Do not group all these cases under the vague label of technical improvement.

For each scenario, record exactly what changes. Translating one approved lesson has different limits from creating new promotional messages every week. Localizing an existing program differs from training a model that can be used in subsequent courses. The teacher should understand this distinction before the source recordings are uploaded.

This distinction also helps with budgeting. If you only need subtitles and an ordinary voice-over translation, there is no reason to create a personal voice model. Choose the simplest method that meets the learning need. The system's additional capabilities are not, by themselves, a reason to increase the volume of data and the school's future obligations.

Rights in a Recording Do Not Mean Control over a Person

For a Ukrainian project, the starting points are the Civil Code, particularly its provisions on filming and the use of images, and copyright legislation. However, the specific AI scenario requires separate analysis: the right to use a lecture and permission to create new statements in its author's name should not be conflated.

Check the chain of rights in the source material. Who filmed the video, who created the presentation, is there third-party music, and do students appear on camera? The teacher's permission does not remove every other restriction. Conversely, a contract with the videographer does not automatically resolve the use of someone's likeness in a new product.

Do not rely on the recording already being publicly available on YouTube. Public availability makes viewing easier, but does not explain which forms of reuse are permitted. Your workflow needs verified source files and documents rather than a random clip downloaded online. If the chain of rights is unclear, start by establishing it.

Describe the Permission in Plain Language

Consider a hypothetical school localizing a project management course. The teacher agrees to Polish dubbing of twenty specific lessons for current and future buyers of that program during an agreed period. That is a clear basis for discussion. Wording covering every possible use in any product does not provide the same transparency.

Define the materials, purpose, languages, channels, audience, duration, and responsible organization. Discuss advertising, public social media, corporate versions, and transfers to contractors separately. A teacher may permit educational dubbing and decline a promotional avatar. This should be a real choice rather than something hidden inside an indivisible package of terms.

For projects subject to the GDPR, the EDPB explains the requirements for consent: it must be freely given, specific, informed, and unambiguous. This does not mean consent is the only possible basis for every instance of processing. The school needs to identify the applicable law and an appropriate basis for the particular operations.

Show the Result before Final Approval

Explaining voice cloning in words is harder than showing a short, controlled example. A teacher may agree to the technology without expecting the system to change intonation, pace, or emotional tone. The demonstration should be created within a previously agreed test, rather than by secretly uploading a recording to make the idea easier to sell.

Choose a neutral passage without sensitive claims for the test. Let the person listen to the result, compare it with the original, and ask questions. Explain the limitations too: incorrect stress, unnatural pauses, a different emotional tone, and inaccurate terminology. Realistic sound does not guarantee an accurate representation of the author's views.

Record what has been approved: the possibility of creating a model, a specific passage, or the rules for future work. Do not interpret a positive comment about the test as permission for unlimited use. After the demonstration, both parties should share an understanding of the next step, who is responsible, and how the finished material will be reviewed.

Check the Service before Transferring Files

Find out which recordings are needed to create the model, where they are processed, how access is restricted, and what happens after the account is deleted. A particular product's terms may vary by plan, region, or enterprise agreement. Do not apply an answer from an old promotional page to the current scenario without checking it.

For example, the ElevenLabs Prohibited Use Policy restricts reproducing a person's voice or likeness without appropriate permission. However, following the service's rules does not replace agreements between the school and the teacher. Nor does the teacher's permission remove the provider's additional requirements for creating and using a particular type of voice model.

Keep the agreed version of the terms and the date they were checked. If the service changes its rules, someone should assess the consequences for materials already created. Do not leave legal and technical information solely in an editor's private correspondence. A year later, the team should still understand the terms under which the voice-over was created and who may use it.

Define Who Owns the Account

A personal avatar should not be created in a contractor's arbitrary account. Otherwise, the school may lose access to the materials, and the teacher may lose the ability to control how their likeness is used. Choose an ownership and delegation arrangement compatible with the service's rules. Resolve access before the first generation, rather than after the contractor's engagement ends.

Keep the list of authorized people short and tied to roles. One employee prepares the text, another checks the content, and a third performs the technical generation. Not everyone needs permission to create any statement they choose. Keep a work log and regularly review access, especially after staff changes. A shared password in a chat does not provide accountable control.

Check external contractors separately. If they receive source recordings or access to the model, their obligations should match the agreed scenario. Prohibit using the materials in their own portfolios without permission. Testing other services also requires a separate decision: a contractor's convenience does not define new purposes for data processing.

Approve the Content, Not Just the Similarity of the Voice

In a synthetic lecture, someone can sound convincing while saying something incorrect. A learning product needs a substantive review of the text, translation, and finished audio track. The author should know who is responsible for facts, formulas, examples, and program changes. A realistic avatar increases trust, so an incorrect phrase may cause greater harm.

For translation, involve someone who understands both the subject and the language. An automatically generated sentence may be grammatically correct yet use an unsuitable professional term or change the strength of a claim. If the teacher does not speak the language, they cannot personally check every nuance. You need an agreed editor and a clear approval process rather than a formal 'approved by the author' label.

For promotional materials, establish a separate review of promises. Do not make a digital copy speak about income guarantees, personal experience, or recommendations the real person never gave. Even with broad permission, the school remains ethically responsible for an appropriate context. Generation should not create a false biography for the teacher.

Clearly Label Synthetic Material

Students should understand that they are encountering AI narration or a digital avatar. Specific labeling obligations depend on the applicable law and method of distribution; check them for your project. As an editorial practice, a short notice next to the material can explain the technology used and how the content is checked, without overwhelming readers with technical details.

Do not give the impression of a live personal message if the video is generated automatically for many people. This is especially important when it contains an assessment, praise, or a recommendation. Students may give the words different weight if they believe the teacher personally reviewed their work. The message format should reflect the actual process.

If the video is hosted on the WeStudy LMS platform, the school must ensure that the material has an accurate description and the agreed access settings. Having a learning platform does not resolve permissions for someone's likeness or the review of synthetic text. Hosting follows editorial and legal approval; it does not replace those procedures.

Agree in Advance on Ending Use

Distinguish stopping new generation, deleting the model, and continuing to use materials already released. These are different operations with different consequences. A school should not promise to instantly erase every copy from the internet if buyers or third-party platforms have already lawfully received some videos. The process should be realistic and legally reviewed.

In the contract and internal instructions, define a contact, verification of the requester's identity, responsible people, and action deadlines for each scenario. What happens to an active course? Can an ordinary voice-over replace the track? Who notifies contractors? Will necessary evidence of previous approvals be retained? The answers depend on legal grounds and the terms of collaboration, rather than the editor's preferences.

Run a small exercise using test material: the team receives a request to stop new use and works through every step. If no one knows where the source files, model, and promotional clips are stored, the process is not ready. It is better to discover this in a ten-second example than after launching a large multilingual library.

If the Digital Copy Says Too Much

Start by stopping distribution of the specific erroneous material in channels you control. Preserve the information needed to investigate the incident: the text, version, approvals, release time, and access group. Do not alter the history in ways that prevent you from understanding the cause. Notify the responsible people, including the teacher, and agree on an appropriate correction.

Distinguish a content error from a breach of permitted use. The former may require editing the lesson. The latter may require reviewing access, contracts, or a contractor's work. If personal data is at risk or there are other legal consequences, involve a qualified professional to determine the necessary notifications and actions. Do not apply one universal template to every situation.

After correcting the problem, change the process that allowed it. For example, separate promotional script approval from educational script approval, add a language review, or restrict generation to finished, approved texts. Risk reduction should address the specific cause rather than rely on the team's promise to be more careful next time.

Translation Can Change the Author's Position

Consider an educational example: a teacher says a certain method sometimes helps under specified conditions. In translation, 'sometimes' and 'under specified conditions' disappear, and the digital voice confidently recommends the method to everyone. A formal grammar check will miss this error. For the author, the strength of the statement has changed; for the student, the practical recommendation has changed. This is why approval should concern the meaning as well as the sound.

Ask the editor to flag statements where limitations, probability, or professional qualifications matter. For particularly difficult passages, you can use a meaning-based restatement in the author's language and a subject expert's review. This does not guarantee perfect localization, but it makes the process more transparent. If the school lacks the resources for this review, it is better to limit the translation or change the format than to distribute unchecked advice in the voice of a recognizable expert.

Separate the Learning Channel from the Advertising Account

The same track can be perceived very differently in a private course and a short advertisement. In learning, the phrase is surrounded by context, examples, and limitations. Promotional editing often retains only the most striking sentence. Course approval should not be treated as automatic editorial approval of such a clip. For advertising, define a separate list of permitted scenarios and someone responsible for the final review.

Take particular care with personalization. A synthetic presenter may address someone by name, but that does not mean the teacher knows them or personally recommends a product to them. Do not add an invented history of interaction to the video. Nor should a recording be moved into advertising for another school, a partner's product, or a public campaign without new approval. To a viewer, a recognizable voice signals endorsement, even if the team regarded it simply as a convenient audio asset.

Keep a Register of Digital Materials

Create a simple list in which every synthetic version has a clear identifier, date, source, language, purpose, and link to its approval. Separately indicate where the material has been published and who can replace it. This need not be a complex system: for a small project, an organized spreadsheet with restricted access is sufficient. The key is to avoid relying on one editor's memory.

The register also helps with ordinary editing. If a definition changes, you need to find every language track and promotional excerpt in which it appears. If the permitted period of use ends, you can identify the relevant channels. Store the documents and source files themselves in an appropriate environment, and keep necessary links and administrative information in the register. Do not accumulate unnecessary personal data merely to make the spreadsheet more comprehensive.

Agree on Compensation and Additional Work

Creating a personal model can change the usual economics of collaboration. Previously, the teacher was paid to record each lesson; now, the school can generate additional materials without filming again. Discuss this openly: what the compensation covers, which volumes are included, whether script reviews are paid, and how new programs are agreed. There is no universally fair rate, but a hidden change to the rules creates conflict.

Account for the author's own work after generation too. Listening to a two-hour course in another language, agreeing terminology, or checking many promotional versions is a separate workload. If the school expects quick approval, agree on a deadline and a format for review materials. Do not force someone to choose between formal approval and disrupting your launch. Quality control takes time, just as traditional filming required preparation.

What to Ask in the First Conversation with the Teacher

Start with the purpose: why does the school need synthetic material, and why is an ordinary recording insufficient? Then let the person state their own boundaries. They may permit translation only, want to check every text, or not want the model used after the collaboration ends. Treat these answers as project conditions rather than obstacles to bypass through broader contract wording.

Agree separately on how to handle new service features. Today, the team creates narration; tomorrow, the provider offers an interactive conversational agent with the same voice. This transition may substantially change the risks and expectations. Requiring renewed approval for a new purpose helps the school develop its product without silently expanding the original permission. After the conversation, prepare a brief description of the decision that the author, producer, lawyer, and technical contractor can all understand in the same way.

A Small Review Package for a Lawyer

Before the consultation, gather a description of the planned operation, the original contracts, a list of services, the audience's countries, and an example of the finished material. Add the proposed limits on use and the process for stopping generation. The professional can then review a specific workflow rather than an abstract question about the legality of artificial intelligence. This also helps reveal contradictions between the author's permission, the service's terms, and the school's promises.

After the review, give the team short instructions describing actions and limits, alongside the full contract. The editor should know which materials may be used, the producer should know when renewed approval is needed, and the administrator should know who has access to the model. A legal arrangement works when it can be followed daily without guesswork or informal exceptions. This provides more practical support for a launch than a general belief that everyone involved has supposedly already agreed.

A Useful Alternative to a Personal Clone

Not every course needs a digital double of a real teacher. You can use ordinary narration, a licensed synthetic voice, a neutral virtual presenter, or a video without a face. These options also have terms of use, but may suit the task better. Discuss the technology choice with the program's author rather than presenting them with a decision already made.

For example, short technical updates may only need a screen demonstration and a clear audio track. An author's introduction may work better as a real message recorded once. To localize a complex lecture, you can retain the original and offer subtitles. A personal clone should be a justified tool rather than a compulsory symbol of the school's modernity.

When updating a course through the WeStudy LMS platform, you can organize access to approved versions of materials according to the project's capabilities and settings. However, the people creating the product decide whose voice is heard and within which limits it is used. Good technical organization begins with a clear agreement between them.

In the finished material, also check the intonation around sensitive wording. An ironic or overly confident tone can change how even a formally correct text is perceived. If the system does not convey the intended meaning properly, record the passage another way. A tool's limitations should not become a reason to attribute an unwanted manner of speaking to a real person.

Sources

1. Civil Code of Ukraine.

2. Law of Ukraine 'On Copyright and Related Rights'.

3. Law of Ukraine 'On Personal Data Protection'.

4. EDPB — Lawfulness of Personal Data Processing.

5. ElevenLabs — Prohibited Use Policy.

6. U.S. Copyright Office — Report on Digital Replicas, 2024.