TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Reward

Part of

Deep Learning Model Evaluation Benchmarks