Panel Discussion on AI, Art, and Creativity

Event: The future of generative visual art workshop @ CVPR 2024 · Duration: 103 min · ▶ Watch on YouTube

Abstract

This workshop explores the intersection of AI, art, and creativity, featuring presentations from leading researchers and practitioners. Speakers delve into the historical context of technology’s impact on art, examine current advancements in controllable visual imagination through generative AI, and discuss the future of video generation. The panel collectively addresses the evolving relationship between human creativity and AI tools, emphasizing the need for intuitive interfaces, precise control, and interdisciplinary collaboration to foster new artistic expressions and navigate the challenges posed by rapidly advancing technologies.

Speakers

  • Aaron Hertzmann — Adobe Research
  • Jia-Bin Huang — Professor@UMD, Research Scientist@Meta
  • Anastasis Germanidis — Co-founder, CTO RunwayML

Talks (4)

  • 00:00:00 — Aaron Hertzmann: When Technology Changes Art
    • Explores historical trends in artistic technologies to understand cognitive biases around change, emphasizing that AI is a tool for art, not an artist, and highlighting patterns of experimentation, style evolution, backlash, and shifting skills.
  • 00:03:18 — Jia-Bin Huang: Controllable Visual Imagination
    • Discusses the evolution of visual art creation from ancient cave paintings to modern AI, focusing on how human creators can achieve precise control over generative models through expressive text-to-image generation with rich text and image-guided texturing.
  • 00:08:08 — Anastasis Germanidis: Video Generation and Controllability
    • Highlights rapid advancements in video generation, focusing on improvements in fidelity, geometric consistency, and identity maintenance, and discusses the importance of dense captioning and interactive tools for artistic control and understanding of physics within generative models.
  • 00:16:00 — Panel Discussion: Panel Discussion on AI, Art, and Creativity
    • Panelists discuss their personal experiences with AI art, the evolving role of generative AI in creative workflows, the importance of iteration and control, and the need for interdisciplinary collaboration to push the boundaries of AI-assisted art.

Key Takeaways

  • AI is primarily a technology that artists use as a tool, not an artist itself, and its impact on art should be viewed through a historical lens of technological change.
  • Controllability and intuitive interfaces are crucial for empowering human creators to leverage generative AI effectively, allowing for precise artistic expression and exploration.
  • The rapid advancement in generative models, particularly in video generation, is enabling new forms of artistic expression and interaction, but also raises challenges regarding ethical use, intellectual property, and the definition of art itself.
  • Interdisciplinary collaboration, involving fields like HCI, psychology, and philosophy, is essential to develop AI tools that truly augment human creativity and address the complex societal implications of these technologies.
  • The future of AI art lies in developing models that understand the underlying ‘rules of the world’ and allow for interactive, real-time control, moving beyond simple image generation to complex, controllable simulations.

Methods / Models / Datasets Mentioned

  • Inceptionism
  • DeepDream
  • Neural Style Transfer
  • CycleGAN
  • Pix2Pix
  • BigGAN
  • StyleGAN
  • DALL-E 3
  • Midjourney
  • Emu (Meta)
  • Imagen (Google)
  • Firefly (Adobe)
  • SDXL (Stability.ai)
  • BLIP
  • DreamBooth Finetuning
  • Score Distillation Sampling (SDS)
  • Variational Score Distillation
  • Gen-2 (RunwayML)
  • Gen-3 Alpha (RunwayML)
  • InstructPix2Pix
  • Masa-Ctrl

Topics

AI in art · Generative models · Artistic technologies history · Cognitive biases in art · Controllable visual imagination · Text-to-image generation · Image-guided texturing · Video generation · Human-AI collaboration · Interdisciplinary research

Notes

Open for commentary — connections to other work, critiques, follow-up reading.