generative media
An image, video, audio, or music that was not recorded by a camera or microphone but generated by artificial intelligence, mostly based on text instructions.
In plain terms
Generative media is a collective term for any image, video, speech, music, or 3D model created by an AI model. No recording is made; instead, the machine "draws" or "voices" the content based on your request.
Imagine an extremely fast illustrator who has already seen hundreds of millions of images and films. Mostly, it does not copy one from what it has seen, but assembles something new from learned patterns, although it rarely happens that it returns a frequently seen image almost unchanged. If you ask for "a vizsla in a spacesuit on the shore of Lake Balaton", it will be ready in a few seconds.
It is useful for having illustrations, advertising videos, background music, or read-out text without expensive filming, a studio, or a graphic designer. Traditional editing programs transform existing recordings, but here the raw material also comes from the machine.
The same capability is also its danger: it can be used to make deceptively realistic fake recordings, even with the faces and voices of real people.
How it works
- Training: the model receives vast amounts of images, videos, or audio recordings along with their descriptions and learns which words correspond to which shapes, colors, movements, and sounds.
- Instruction: you specify what you want. This is usually text (prompt), but it can also be a sample image, sketch, or audio clip.
- Generation: most image and video generation models start from random 'noise' and clean it up in many small steps until content matching your description emerges. This is called diffusion. Speech and music models, as well as some newer image models, build content piece by piece using a different method.
- Refinement: you can improve the result with further instructions, for example requesting a different background or slower camera movement, and choose from multiple variations.
- Labeling: serious services embed invisible watermarks or provenance data in files to make machine origin detectable later.
An everyday example
You scroll through a social media site and see a touching video of a little boy hugging a bear cub in the forest. The recording never happened: someone had it created by a video generator from a single sentence. The same technology family can read your audiobook aloud and create the background for an online store's product photo.
Who is affected
A common misconception
Limits and risks
- It can also be used for deception: fake videos, scams using voice clones, or defamatory images can be created in minutes.
- The model does not know reality, only patterns, so it can generate plausible but false details, such as incorrect captions, non-existent buildings or impossible movements.
- It is legally unsettled what rights creators of works used for training are entitled to, and in many countries purely machine-generated content does not receive copyright protection.
- Recognition programs and watermarks are not infallible: markings can be removed, and the detector may classify a real recording as fake.
Similar, but not the same
- deepfake – A deepfake is a type of generative media that appears deceptively real and usually imitates the face or voice of a real person. Most generative media is not like this and is not created to deceive.
- generative AI – Generative AI is the broader technology that produces text and code. Generative media is the visual and audio part of this.
- CGI (computer-generated imagery) – CGI is built by artists using professional software, element by element, while generative media is produced by an AI model from a single instruction. The two methods are now often used together.
Related concepts
generative AI, artificial intelligence, szintetikus tartalom
A brief history
One starting point of today's generative media is the Generative Adversarial Network (GAN) introduced in 2014. The first version produced only small, blurry images, while improved versions by the late 2010s could generate convincing artificial portraits. It reached the general public in 2022 with the launch of text-to-image services anyone could try, such as Midjourney, Stable Diffusion, and DALL·E 2.
Where things stand
By the end of 2026, image, audio and music generation had become an everyday tool, built into phones, office and design software, and short videos produced together with sound are spreading en masse. The transparency rules of the EU AI Act, applicable in stages from August 2026, require machine-generated content to be labelled and deepfakes to be disclosed, but practical implementation and the copyright disputes are not yet settled.
In the news
- Meta makes advertisers solely responsible for its AI changes to ads; new API features arriveNov 2, 2026 · Marketing
- Etsy restricts AI-generated content to curb seller slopOct 23, 2026 · Marketing
- Hungarian universities' AI rules: ELTE criterion course, MATE plagiarism rule, PTE interviewOct 19, 2026 · Education & Family
- Axios: AI slop websites up fivefold as GEO optimization spreadsOct 17, 2026 · Marketing
- AI ad labelling: responsibility stays with the advertiser, while the EU requires transparencyOct 12, 2026 · Marketing
- Stack Overflow relaunches Developer Story to filter AI-noisy recruitmentOct 12, 2026 · Code
- AI Act transparency rules: labelling AI content and deepfakesOct 12, 2026 · Jog
- Podcasts address employment-law and compliance questions on workplace AI useOct 6, 2026 · Jog
Also known as
generative media · generative media · AI-generated content · AI-generated content · synthetic media · synthetic media · AI-generated content · AIGC
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.