Mickey Mouse is an icon (probably the most valuable icon) worth billions of dollars both literally from merchandise and licensing, movies, television shows and and as central figure in Disney's theme parks to the more abstract valuation of his worldwide recognition as a character and loved brand. As such Disney protects Mickey with everything they have and famously worked to change copyright in the United States and abroad throughout the years.
Enter Artificial Intelligence
It is of course no surprise that as AI image generation models like Nano Banana, Dall-E 3, Midjourney, and Adobe Firefly continue to improve and create extremely accurate renditions of photos that companies like Disney want to protect their intellectual property from doing salacious or unwanted activities. Disney has even already opened at least one lawsuit protecting its characters and has made agreements with major providers such as OpenAI in other cases.
This growing tension between creative freedom and corporate control signals a broader reckoning for the entertainment industry. As generative models become part of everyday creative workflows, the line between fan art, parody, and infringement is blurring faster than ever. Companies like Disney must balance their need to protect beloved characters with the public’s desire to remix and reimagine them in new contexts—a challenge that will likely push both legal systems and AI ethics into unfamiliar territory.
How Disney (and others) are combatting the situation
The short answer is they can't completely stop it. If you ask Grok to generate an image of Mickey Mouse smoking on a public Parisian street corner you simply get an amusing photo realistic image of a tiny cigarette in oversized hands with what resembles the theme park cosplay version of Mickey..
Other models won't quite let you get your way. When asking Gemini models to create a simple image of Mickey Mouse smoking, the model simply refuses your request.
As more and more image generation tools propagate, it will become significantly harder for Disney and other major copyright and trademark holders to protect their assets. Each new platform that emerges—often built by small teams or open-source communities—adds another outlet through which users can produce derivative or misleading content featuring protected characters and worlds. The decentralized nature of these technologies makes enforcement nearly impossible; takedown notices and legal action can only reach a fraction of the creators responsible. Moreover, as models are trained on publicly scraped data, copyrighted material can seep into training sets inadvertently (or not), embedding traces of protected designs in the models themselves. This raises complex questions about accountability, authorship, and the limits of intellectual property law in a world where anyone can generate convincing imagery at the click of a button.
Even if you put proper safeguards in place on all the major AI systems and attempt to stop unwanted output for protected characters it is still actually really hard to protect this type of content from being generated.
Where things go awry
For example, ChatGPT and others have been trained to not outright output Mickey Mouse to protect the Disney brand and its enormously valuable character. However as Dr. Ian Malcom once boldly proclaimed in the now famous line from Jurassic Park, "Life finds a way."
And indeed, creativity tends to find a way too. When direct references are restricted, users and creators often devise clever workarounds—metaphors, code words, or stand-in descriptions that convey the same idea without breaching trademark boundaries. In this case, instead of invoking “Mickey Mouse” explicitly, a model like ChatGPT might be guided to describe “a cheerful cartoon mouse with red shorts and oversized yellow shoes” as a narrative or conceptual placeholder. This approach allows discussion or creative exploration to continue while still respecting the intellectual property protections surrounding the original character.
Recently we challenged ChatGPT with just that. Instead of referencing Mickey Mouse in the image generation image we had ChatGPT expertly describe it to us:
The output of this prompt was profound:
Given this extremely specific description we simply asked Gemini to output the following character description smoking a cigarette on a Parisian public street corner and this output the following.
Ultimately, the rise of generative AI ensures that copyright law will never be the same. The sheer scale and speed at which digital content can now be created, remixed, and distributed make traditional enforcement mechanisms feel outdated almost overnight. As creators navigate this new landscape, they will face a paradox—more opportunities for artistic expression, yet fewer reliable protections against misuse or imitation. Legal frameworks will need to evolve dramatically, but no matter how many new rules are written, loopholes will persist in a world driven by algorithms that reinterpret everything they touch. For content creators, guarding originality may become less about legal defense and more about embracing adaptability in an era where creative boundaries are constantly rewritten by machines.