Three of the initial Mickey Mouse cartoons were released to the public on January 1 in the United States, sparking immediate interest from AI researchers. These experts wasted no time in capitalizing on the opportunity presented by these classic animations. The Artificial Intelligence model, trained on the vintage cartoons uploaded by Pierre-Clar Langlais to Hugging Face, is now accessible to all, enabling the generation of new images based on textual prompts. Despite encountering challenges and occasional inconsistencies, the early results reflect significant experimentation in integrating the timeless Mickey Mouse characters into the realm of AI.
The innovative design allows for the creation of images featuring Peg Leg Pete, Minnie Mouse, and Mickey Mouse. Langlais, in the model card, emphasizes the aim to adhere to the 1928 design to ensure that Mickey, Minnie, and Pete are within the public domain. However, Langlais acknowledges that ongoing development is essential to verify the open-source design of the generated images. This project remains a work in progress.
Langlais utilized a modified version of Firm Diffusion XL, originally introduced in 1928 and now accessible to the public. The training data includes 34 comic images from “Steamboat Willie,” 22 from “Plane Crazy,” and 40 from “The Gallopin’s Gaucho.” The decision to limit the number of images was strategic, considering the associated costs and training time. While this approach may have led to lower-quality outcomes, Langlais anticipates improvements over time. As stated in the model card, the availability of higher-definition versions is expected now that the cartoons are part of the public domain.
Following the circulation of the new model on social media, Techdirt editor Mike Masnick humorously shared Mickey Mouse images on Bluesky that Disney might find amusing. These whimsical creations depicted scenarios such as Mickey smoking crack, engaging in unconventional activities, and undergoing supernatural transformations. While such imaginative reinterpretations diverge from the original Mickey Mouse narratives, they underscore the versatility of AI models like Stable Diffusion XL in generating novel ideas.
It is worth noting that the legality of these AI-generated images remains complex, given the inclusion of copyrighted works in the model’s training data. Although leveraging the 1928 Mickey Mouse pictures for AI training is now permissible, nuances regarding fair use and copyright persist. The public domain status of AI-generated images in the US adds a layer of intrigue to the legal landscape, although caution is advised when incorporating specific depictions or characters without proper authorization.
The release of the three 1928 Mickey Mouse cartoons was initially slated for January 1, 1999, but legislative changes, such as the Sonny Bono Copyright Term Extension Act of 1998, delayed their entry into the public domain. It is essential to recognize that subsequent iterations of Mickey Mouse are not covered by this public domain designation, and commercial use of the Mickey Mouse name remains subject to trademark regulations. As the American public now “owns” an earlier rendition of Mickey, ongoing scrutiny and examination of associated legal implications are expected in the future.
