What is synthetic media?

Synthetic media is video, audio, images or text that is generated or substantially altered by artificial intelligence instead of being captured directly from the real world. Examples include AI avatars, synthetic and cloned voices, AI-generated images, and translated videos with regenerated lip movements.

Reviewed by Clark Quinn, Chief Learning Strategist Updated October 28, 2026

Example

A manufacturing company’s L&D team rebuilds its plant safety induction. Instead of filming, they use an AI avatar to present each module, a synthetic voice for the narration and AI-generated illustrations of common hazards. They translate the course into three languages, and the avatar’s lip movements are regenerated to match each one.

Before launch, the team adds a line to the course introduction saying the presenter is AI-generated. A safety engineer stays the named reviewer of every script, and the team files the signed consent of the plant manager whose custom avatar opens the course. All of this is synthetic media, and the team produced it openly and with permission.

How it works

Synthetic media comes from generative AI models trained on large collections of real examples, such as recorded speech, photos or video. A model learns the patterns in that data and produces new content that follows them, guided by an input such as a script, a prompt or a reference recording.

Main types

• Synthetic video: AI avatars and presenters, generated clips and scenes, and real video with altered faces or lip movements.
• Synthetic audio: text-to-speech voices, cloned voices, and generated music or sound effects.
• Synthetic images: illustrations, photos and graphics generated from a text prompt.
• Synthetic text: scripts, summaries and quiz questions written by large language models.

Common techniques

• Speech synthesis and voice cloning turn written text into speech, either in a stock voice or in a voice modeled on a real person.
• Avatar and lip sync models animate a face so it matches new audio.
• Diffusion models generate images and video by starting from random noise and refining it step by step toward what a prompt describes.
• Large language models generate, summarize and translate text.

A single training video often combines several of these: a language model drafts the script, speech synthesis voices it and an avatar model presents it.

Synthetic media vs deepfakes

A deepfake is a type of synthetic media that imitates a real person, usually without their consent, to make it look as if they said or did something they did not. Deepfakes are used for fraud, harassment and misinformation, and they drive much of the public concern about synthetic media.

The underlying technology overlaps. What separates responsible synthetic media from a deepfake is:

• Consent: the people whose face or voice is used agreed to it.
• Disclosure: viewers are told, or can easily tell, that the content is AI-generated.
• Intent: the content informs or teaches instead of deceiving.

An AI presenter in a compliance course meets all three tests. A fake video call from a “CEO” asking finance to approve an urgent payment meets none of them.

Responsible use

• Get written consent: before creating an avatar or voice of any real person, agree in writing how it can be used and for how long. A custom AI avatar of an employee needs the same care as any other use of their personal data.
• Disclose AI-generated content: tell learners and viewers, for example with a note in the video or the course description.
• Keep a human reviewer: someone accountable checks every script and output for accuracy, because AI can produce confident mistakes.
• Protect access: limit who can generate content with custom avatars and cloned voices.
• Choose vendors carefully: look for consent checks, content moderation and clear policies on how your data is stored and used.

Why it matters for L&D teams

• It is already in your toolkit. AI voiceovers, avatars and translated videos are all synthetic media, so the term covers everyday training production.
• Transparency protects trust. Telling learners up front that a presenter is AI-generated avoids the sense of being misled later.
• Consent protects your people. Clear consent and access rules let employees take part in custom avatars without losing control of their likeness.
• It is also a security topic. Staff need to recognize deepfakes, such as cloned voices used in payment fraud, which makes synthetic media a subject for security awareness training as well as a production method.