Generative Models

When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators
We show that newer text-to-image models are progressively worse as training data generators, despite better visual quality, because they collapse to a narrow aesthetic-centric distribution that diverges from real data.
Krzysztof Adamkiewicz, Brian Bernhard Moser, Stanislav Frolov, Tobias Christian Nauen, Federico Raue, Andreas Dengel
PDF Cite
When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators