AI children's books, body horror edition
The author highlights how AI-generated children's books produce visually repetitive and unsettling imagery, using body horror-like examples to illustrate the lack of originality in AI content.
Background
- The article is by Michał "lcamtuf" Zalewski, a well-known security researcher, author, and technologist who often writes about AI, technology, and society on his Substack.
- It critiques the homogeneity of AI-generated content — the tendency of models to produce highly similar outputs regardless of the prompt, especially in image generation.
- The "body horror" reference points to a specific example: a series of AI-generated children's book illustrations that are visually creepy, anatomically wrong, and uncanny in a way that reveals the model's limitations, not creative intent.
- This fits into a broader ongoing debate about generative AI: whether it can produce truly novel work or merely remixes of its training data, and whether its outputs are "good enough" to displace human creators.
- The post is part of a larger pattern where critics argue that AI art collapses into a narrow aesthetic — smooth, plastic-like, and oddly similar across different prompts and platforms.