Forum
Licensed or Open? Two Roads for Creative AI Opened on the Same Day

On 25 November, two announcements described two different futures for creative AI.
In music, Warner Music Group settled its copyright lawsuit against Suno, the AI song generator, and announced a partnership. Suno will launch new, licensed models in 2026 that are meant to replace its current ones, and downloads of audio will be limited to paid accounts. Neither side disclosed financial terms; both framed the deal as a way to compensate and protect artists and songwriters.
In images, Black Forest Labs, the German company behind the FLUX models, released FLUX.2. It comes in several versions, including one with open weights that anyone can download and run, and the company describes it as production-ready, with support for multiple reference images and output up to four megapixels.
One road leads toward models built on licensed catalogs, controlled by agreements between rights holders and AI companies. The other leads toward open models that run on your own hardware under published license terms. Both will shape the tools designers and artists use next year.
The licensed road
The Warner–Suno deal is part of a broader shift in music from litigation toward licensing. A month earlier, on 29 October, Universal Music Group settled its own case against Udio, another AI music service, and the two companies agreed to build a licensed creation platform for launch in 2026, with tracks kept inside the service in the meantime. The logic is straightforward: if models are trained on catalogs with permission and payment, the legal cloud over their outputs thins, and rights holders get a share of a new market.
For creators using such tools, licensed models offer clearer ground. The trade-offs are also clear. Licensed systems tend to come with tighter controls on what can be generated, how outputs can be downloaded and used, and which artists’ names or styles can be invoked. That is a feature for some users and a limitation for others.
The open road
Open-weight image models offer something different: control. A studio can run them locally, fine-tune them on its own material, keep client work off third-party servers and inspect what the model does. For many designers, that independence is worth a great deal.
But “open weights” does not mean “anything goes”. Each release has a license, and terms differ between versions, including on commercial use. Running a model yourself also moves responsibility to you. Questions about what it was trained on, and what you may do with its outputs, do not disappear because the file sits on your own machine.
Reading the license is now part of the craft
Designers learned long ago to check font licenses and image rights. Model licenses belong on the same list.
- Commercial use. Can outputs be used in paid client work, and are there limits based on company size or revenue?
- Training data statements. What does the provider say about the data behind the model, and is that enough for your clients?
- Indemnity. Does anyone stand behind the outputs if a claim arises?
- Restrictions on content. Are there prohibited uses that your project might touch?
- Version differences. Is the version you are using under the same terms as the one in the press release?
What this means for award entrants
Competitions that recognize AI work increasingly expect entrants to know their tools. The International AI Award asks entrants which models and versions they used, why they chose them, and, for code-based work, to state the version of any open-source model and the modifications made. Choosing a model is a creative and ethical decision, and juries will treat it as one.
Neither road is free
It is tempting to frame licensed models as safe but restrictive, and open models as free but risky. The reality is more practical. Both require reading terms, keeping records and making choices that a client could understand. The skill that matters is not loyalty to one camp. It is knowing which road a given project needs, and being able to explain why you took it.



