FP: AI for Experimental Animation

"Shred" experiment, Eaven Harrington

The final project for the "AI for Experimental Animation" class involves creating a short experimental animation using one of the tools we explored throughout the course. In past projects, students have used a variety of tools, from early image synthesis software like Pics to Pics, to Deep Dream, which encourages computers to "hallucinate" by finding imagery in noise and producing synthetic images. This process can reveal the embedded biases of the AI. Stable Diffusion has become particularly popular, especially this year in 2024, and I hope you enjoy these explorations.

Students of experimental animation are often interested in streamlining their workflows. Until this year, many felt that the outputs from these AI techniques were not useful in their personal work due to the overwhelming texture, style, and artifacts introduced by the AI. However, 2024 marks the first time that nearly half of the students found these tools practical for their projects. They reported that the AI-generated images could serve as a foundation for further rotoscoping or structure for their final works. This development represents a significant shift in how these tools are perceived and utilized.

As you review these pieces, you'll notice that the students are keen to explore the specifics of the medium. They emphasize the AI's "failures" and noise—the idiosyncrasies of the technology—rather than seeking to produce traditional, polished images. Their goal is to open up the "black box" of AI, investigate its contents, and openly discuss its characteristics. I encourage this approach, as it fosters a deeper understanding and a more innovative use of the technology.

JEONGEUN LEE

AN BINH TAT

Binh used nonsense prompts to access the preoccupations and primal mind of a Stable Diffusion model. He is interested in foregrounding the failures to uncover the true texture of this process, and it looks like he is on to something.

IAN CHO

Ian surprised me with this critique of classical "women in trouble." She used the lateral associations common in Stable Diffusion generators to critique these Gothic portraits with cut-in imagery.

CATHERINE JEONG

Cathy made a LoRA from a stop motion animation of this character. This LoRA is able to match these analog puppets. This is difficult to achieve with AI tools!