We used cutting-edge and experimental AI image generation techniques to create all the imagery for this show concept. Using deep AI model training, we brought the creators’ vision to life through holistic visual storytelling, maintaining specific details and consistent styles across characters, faces, clothing, and settings.
PROBLEM
Doji on the Hill is a modern animated TV show concept with themes around financial literacy, Web3 and the HBCU college experience. They commissioned us to use AI to create consistent characters and settings based on real people with specific facial and clothing details, all rendered in an overall style similar to the Boondocks series. Fulfilling all these requirements was a feat that could not be accomplished with mainstream AI image generation tools, eg. MidJourney and Dall-E.
SOLUTION
We utilized bleeding-edge open source and experimental techniques to train and develop bespoke AI models that integrated multiple key elements into the final characters such as consistent illustration style across multiple specific actors, apparel and even environments.
The renders were of such high quality that they were virtually indistinguishable from final keyframe art previously only achievable through labor-intensive and costly manual illustrations.
We were even able to create AI models of 2 specific Invisible Friends from a high-end NFT collection that are characters in the show. Getting the AI to understand the essence of an invisible person was challenging, but we turned the 2D art into a 3D model in order to train the AI on the full figure.
Over the course of our research and development phase, we discovered pain points and room for improvement in the AI art generation process and product landscape, insights that led us to developing our in-house AI art generation tool, Think Diffusion.
