Forum
Licensed Characters and Visible Labels: How December Reframed AI Imagery

Six months ago, Disney and Universal sued Midjourney, accusing the image service of turning their characters into a “bottomless pit of plagiarism”. On 11 December, Disney announced a very different relationship with another AI company. It will invest $1 billion in OpenAI and license more than 200 characters from Disney, Pixar, Marvel and Star Wars for use in Sora, OpenAI’s video app, under a three-year agreement. Users are expected to be able to create videos with those characters starting next year. The deal does not include the likenesses or voices of actors.
Six days later, on 17 December, the European Commission published the first draft of a code of practice on the transparency of AI-generated content. It is meant to help providers and users meet obligations under the EU AI Act that are due to apply in August 2026: marking AI outputs in machine-readable form, and clearly labelling deepfakes and certain AI-generated text.
Together, the two announcements mark a shift. AI imagery is moving from a question of what is technically possible to a question of what is permitted, and how that permission is shown.
From lawsuit to license
The Disney deal does not resolve the arguments about training data. It does show what the largest rights holders want: control, payment and clear boundaries. Characters are licensed; actors’ faces and voices are not. That distinction mirrors one that designers should already be making in their own work between creative properties and people.
For the creative industries, licensing deals of this size also set expectations. When fans can legally generate a video with a famous character in an official app, the line between sanctioned and unsanctioned use becomes clearer, and unsanctioned use becomes easier to challenge.
The design problem inside the code
The EU draft code is technical, but at its heart is a design question: how should people be told that what they are seeing was made or altered by AI?
Machine-readable marks address part of it, allowing platforms and tools to detect generated content. The other part is human-readable labelling, and that is where designers come in. A label that is too small is ignored; one that is too loud becomes noise; one that appears inconsistently across platforms teaches people nothing. The draft is open for feedback until late January, and further versions are expected in the first half of next year.
Several design principles seem likely to matter, whatever the final text says:
- Consistency. A shared visual vocabulary for AI labels would help audiences far more than each platform inventing its own.
- Proportion. Labels should match the degree of manipulation, distinguishing a fully synthetic scene from a lightly retouched photograph.
- Persistence. Labels and marks need to survive cropping, re-encoding and reposting.
- Clarity over alarm. The goal is information, not stigma.
What studios and freelancers can do now
Designers working with AI imagery do not need to wait for 2026 to adopt good habits.
- Keep licensed and unlicensed material clearly separated in your pipeline.
- Record which elements of a deliverable were generated, edited or captured.
- Agree with clients, in writing, how AI use will be disclosed in published work.
- Test how your own labels and credits look at the sizes and formats where audiences will actually see them.
The view from competitions
Awards are already a laboratory for disclosure. The International AI Award asks entrants to describe the models, prompts and ethical choices behind their work, and its judging criteria include transparency. As public labelling rules arrive, competitions can help develop the conventions, showing how process can be explained clearly without diminishing the work.
Looking ahead
Next year will bring licensed character libraries to consumer apps and labelling duties to European platforms. Both will change what audiences expect to know about the images in front of them. Designers can treat that as a compliance burden or as a brief. The second option is better: making AI’s role legible is a design problem, and design is the discipline best placed to solve it.



