Sell images, video and audio to AI

The legally simple yes

Most of the discussion around AI training data is about personal data and confidentiality. Content and media archives are often a different case. If you produced the material yourselves, or bought the rights to it, the question of who may license it is usually clear. And published content generally contains little personal data compared with email or support tickets.

That is why this is often the fastest way to sell images, video and audio to AI – or rather, to license them. You don't hand over the archive. You grant a limited right to use it for training.

What content is interesting?

• Published text – articles, reports, news archives, web content and editorial material.
• Books and manuals – non-fiction, handbooks, instructions and technical documentation.
• Images – image libraries, product photos and photo archives with descriptions or metadata.
• Video – training videos, recorded lectures, documentaries and production archives.
• Audio and podcasts – especially speech in Nordic and other smaller European languages, which is rare in global datasets.
• Source code – your own codebases and documentation where you own the rights.
• Structured datasets – tables, catalogs and compilations you have built up.

Why Nordic and European content is in demand

Smaller European languages make up a small share of the data large AI models are trained on. AI Sweden built GPT-SW3 on a Nordic corpus, and OpenEuroLLM has had to generate synthetic data because of the shortage. In Norway, the National Library found that licensed books and newspapers made Norwegian models better.

The rights question is moving too. The Swedish music rights organization Stim has launched what it calls the world's first collective AI license for music, and in Sweden a national coordinator for Swedish language models is also looking at compensation for rights holders. Read more about what makes data valuable in our overview of AI training data.

The text and data mining opt-out

Under the EU's Digital Single Market (DSM) Directive (Directive (EU) 2019/790, Article 4), as implemented in each member state, text and data mining is permitted – unless the rights holder has reserved those rights in an appropriate way. For material published online, that in practice means a machine-readable reservation.

The EU AI Act (Article 53) also requires providers of general-purpose AI models to respect such reservations and publish a summary of the content they trained on.

The combination is what turns your archive into an asset:

1. A reservation closes off free text and data mining.
2. A license opens paid, limited access for the AI companies that want to use the material.

How you word the reservation is something to check with your lawyer. We help with the licensing part.

How a content license works

The starting point is a non-exclusive license with clear limits on scope, time and use. So you can license the same archive to several buyers, and the buyer may not resell the material. More about the terms on our page about data licensing.

How it works:

1. You describe the material in our form – type of content, approximate volume and who owns the rights. No files needed.
2. We assess rights, any personal data and demand, and get back to you within 48 hours on business days.
3. You approve a concrete offer. Nothing is shared until a non-disclosure agreement and an accepted offer are in place.
4. Delivery and payment take place according to the agreement, after any de-identification.

Our documentation also helps the buyer meet the AI Act's transparency requirements, which makes your material easier to use.

What you need to check

Even the simple yes has a few points to go through:

• Chain of rights – do you own the material yourselves, or are there freelancers, agencies or licensed content in the archive?
• People in images and audio – portraits, interviews and recorded conversations can contain personal data.
• Previous agreements – some agreements with creators or distributors may limit new uses.

If we find something that means parts of the archive shouldn't be licensed, we say so early. This page is general information, not legal advice.