Generative AI technology is making significant inroads into the animation industry, offering innovative solutions and presenting new challenges. Independent studios such as Asteria Film, Invisible Universe, and Toonstar are at the forefront of this transformation, exploring ways to integrate generative AI into their animated content workflows. These studios, as discussed in a previous piece by VIP+, are gradually introducing AI tools to expedite their production processes while remaining focused on originality and artistry.
A crucial aspect of this AI adoption is the process known as "fine-tuning." This technique involves customising a pre-trained image or video generation model using a curated selection of original artworks to create a unique model tailored to a specific character, franchise, or artistic style. This allows studios to maintain visual consistency and integrate their unique intellectual property (IP) into their projects. By using this fine-tuning method, studios aim to enhance artistic output while speeding up the creation process.
"In-house artists or animators develop a 'core set' of original concept art representative of the original character or project," explained an industry source. These art assets become the dataset for fine-tuning any foundation image or video model, such as Stable Diffusion. The resulting customised model serves not only to maintain the studio's character integrity but also to facilitate extensive content creation. For instance, Invisible Universe has integrated custom models for each of its original characters into its content creation workflow, aiding in the rapid generation of material for social media platforms.
Tricia Biggio, CEO of Invisible Universe, elaborated on the significance of fine-tuning: "For us as IP holders, [pretrained] models didn’t know our characters. They were turning them into Pixar or DreamWorks characters. We’ve [customized models] trained on data we own for IPs that we own." This critical distinction highlights the advantages for studios looking to remain true to their creative visions.
Asteria Film echoes this sentiment in their production strategies, where fine-tuning has become an invaluable tool in their workflow. Paul Trillo, a filmmaker and director participating in Asteria's projects, described a recent 2D animation undertaking, stating: "Our artists build out the world themselves and define the style. For one project, our artists drew the main character from every single pose and angle... Then we can go and make a whole city out of that, and it retains the artist’s style.” This reveals how AI can complement rather than overshadow traditional artistry, allowing for more efficient world-building while preserving the unique characteristics of their projects.
A frequently discussed aspect of custom model training involves creating LoRAs, short for low-rank adaptation. This method enables studios to tailor their models further by applying distinct sets for specific projects or characters. Eric Shamlin, CEO of AI-centric studio Secret Level, stated that “every show, movie or brand is going to have their own LoRA,” indicating a trend towards highly personalised content production.
Despite these advancements, the integration of generative AI in animation does not come without risks. Legal and ethical dilemmas loom large, particularly regarding copyright infringement. Sources convey that, although studios focus on developing their own assets, the foundational models for their training have often been built using unlicensed material, leading to potential legal entanglements.
In response to these issues, Asteria has announced plans to utilise Marey, an AI image and video model developed by Moonvalley, which is trained exclusively on licensed and purchased data. Paul Trillo noted, “No scraped data will be part of the pipeline once that becomes available,” indicating their commitment to addressing copyright concerns.
The broader implications of these technologies extend to the artist community within Hollywood. The integration of AI in creative processes raises questions about talent exploitation, with some artists voicing concerns over potential job decreases. Following recent layoffs and a declining trend in artist compensation, industry sources suggest that the impacts of generative AI are becoming increasingly evident, though not directly accountable. Notably, the Animation Guild recently ratified an agreement that did not secure restrictions on training AI models using artist-submitted assets, raising additional concerns about the future landscape for artists and animators.
As independent studios push the boundaries of what generative AI can do for animation, the evolving relationship between technology and creativity continues to shape industry practices, raising vital questions about the future of artistic expression, copyright, and the livelihood of artists within the sector. The conversation is expected to develop as more studios adopt these technologies and the industry pays closer attention to the ethical implications interconnected with AI advancements.
Source: Noah Wire Services