Logo_hovers_2024
Title
Svjetski festival animiranog filma /
3. do 8. lipnja 2024.
Svjetski festival animiranog filma / 3. do 8. lipnja 2024.
hr | en
14_andy_buchanan

ANIMAFEST PRO | ANIMAFEST SCANNER XI | Panel 4: Authorship In/Of AI Animation

The End of the Pipeline - Feature Engineering in Animation Pre-production - Andy Buchanan (United States)

PANEL 4 - AUTHORSHIP IN/OF AI ANIMATION
05/06 SRI 14:35-15:05 KIC

GenAI animation technology advances with stunning speed. Future genAI tools may deliver a fully controllable AI-animation production toolset, but for now, artists often find themselves wrestling with workflows that can be unpredictable and unsuited for some production tasks. The long-established instrumental modularity of the animation production ‘pipeline’ (1) - especially in the pre-production phase - provides opportunities for targeted AI interventions on specific visual tasks. Implementing these tools requires a certain approach to animation based ‘feature engineering’ (2). This term refers to the identification, extraction and use of the most salient data features within a process. In this presentation, the data in question is the utility information embedded within pre-production elements such as storyboards or concept art that are necessary for downstream production tasks. For example, storyboards contain a wealth of diagrammatic information about screen direction, framing and composition, but may appear as very simple sketches in terms of their graphical style. For this reason, we cannot merely train generative systems to 'produce storyboards', as we may produce images that look very much like storyboards, but are not able to function as storyboards. This theme persists across many 'functional images' in the pre-production phase such as design references, scene layouts, lighting diagrams, etc. This presentation explores the creation of functional images for animation pre-production, showing the processes and tools involved: using image constraint systems, human-in-the-loop iteration, camera motion controls and the possibilities for future integration of pre-production to final output systems. The theoretical contribution of the presentation is in three key topics: 1. The ability for generative AI to create ‘functional images’ which have downstream informational content beyond their appearance, and the new creative roles this requires (such as animation feature engineers), 2. The broader implication of these changes for the evolving structure of animation pipelines, including the potential dissolution of the ‘pre-production’ phase as a discrete set of tasks and skills, 3. The ethical challenges of feature engineering in this context, as limited or biased content often arises from dataset selection, model training and finetuning, and prompt engineering.

Andy Buchanan is an animation artist and scholar - his published research and internationally exhibited works lie primarily in the area of experimental digital animation, and reflect an obsession with visual metamorphosis and animation philosophy. Current projects focus on the use of generative algorithms and models for animation production, especially where these new technologies offer new visual phenomena and give rise to new forms of thinking, modes of existence and artistic practices.