The reasons behind the oddities in AI-generated food images
Experts analyze why food images generated by AI models appear unsettling, citing the models' lack of understanding of the physical world and our own evolutionary biological responses.
The use of AI-powered image generation tools in food marketing is sparking debate due to frequently encountered unrealistic and unappetizing images. Experts state that this is rooted in the technical inability of diffusion models to maintain structural integrity and the evolutionary responses of human psychology toward food.
Technical limitations and structural errors
Diffusion models struggle to maintain structural integrity, especially in fine and continuous shapes, during the process of generating images by removing noise. This causes foods like noodles or pastries to appear in tangled, unrecognizable forms.
Because AI lacks a conceptual understanding of what food is, it applies textures incorrectly. It treats food like architectural materials, resulting in contextually disconnected and visually inconsistent outcomes.
Training data and model quality
The density of strange internet images in training data directly affects the quality of the generated results. Furthermore, model collapse, which occurs when AI is trained on its own generated outputs, leads to a further degradation of visual quality over time.
Evolutionary responses and perceptual discomfort
Humans have developed a defense mechanism throughout the evolutionary process to detect signs of spoilage or parasites in food. Consequently, artificial objects generated by AI that resemble worms, holes, or unusual colors trigger an instinctive feeling of disgust in humans.