A blog related to original AI artworks. I make AI artworks using coined words and emoji as prompts.

Poster Presentation at Nicograph 2024

I gave a presentation titled “Using Personal Names or Pen Names as Prompts to Unleash the ‘Artistic Talent’ of Generative AI” at Nicograph 2024, a conference organized by the Society for Art and Science. The presentation covered the Kana and Kanabstracd artworks featured on this site. A summary of the presentation, the poster content, and related materials are available on the paper site. Here are photos of the poster:

I submitted a full paper when applying, but it was not accepted as such and was instead handled as a poster presentation. I also submitted an “art paper” to Siggraph Asia 2024, but it was rejected.

During the poster session, the most insightful discussion was with Dr. (Prof.) Akihiko Shirai from AICU/Digital Hollywood University. We primarily discussed whether AI artworks like those on this site could be considered creative works and whether they are eligible for copyright protection. He emphasized the need for theoretical preparation to support copyright claims, which is a topic worth considering.

We also discussed the reproducibility of images created with Stable Diffusion. Previously, I understood that perfect reproducibility was impossible. However, it turns out that fixing the primary random seed and the sampler seed in Stable Diffusion can achieve reproducibility. While earlier experiments on openart.ai concluded that reproducibility was not achievable, re-testing on the platform revealed that fixing the seed in the interface and selecting the appropriate sampler can produce identical results. It seems this improvement is due to enhancements in the openart.ai interface.

However, fixing the seed in this interface lowers the efficiency of image generation, and the seed of downloaded images must be recorded somehow, increasing the workload. For this reason, I decided to continue generating images without worrying about the seed, as I have done so far.