Autonomous language-image generation loops converge to generic visual motifs.

Autonomous AI-to-AI creative systems promise new frontiers in machine creativity, yet we show that they systematically converge toward generic outputs. We built iterative feedback loops between Stable Diffusion XL (SDXL; image generation) and Large Language and Vision Assistant (LLaVA; image description), forming autonomous text → image → text → image cycles. Across 700 trajectories with diverse prompts and 7 temperature settings over 100 iterations, all runs converged to nearly identical visuals-what we term "visual elevator music." Quantitative analysis revealed just 12 dominant motifs with commercially safe aesthetics, such as stormy lighthouses and palatial interiors. This convergence persisted across model pairs, indicating structural limits in cross-modal AI creativity. The effect mirrors human cultural transmission, where iterated learning amplifies cognitive biases, but here, diversity collapses entirely as AI loops gravitate to high-probability attractors in training data. Our findings expose hidden homogenizing tendencies in current architectures and underscore the need for anti-convergence mechanisms and sustained human-AI interplay to preserve creative diversity.