Tag
Amazon and UCF researchers introduce SatisDive, a training-free inference-time method that formulates text-to-image generation as satisficing—requiring each candidate to meet a reward floor and the batch to meet a diversity cutoff—showing that varying the reward floor traces a Pareto frontier between worst-candidate reward and batch diversity, and that SatisDive Pareto-dominates FK steering on Pick-a-Pic with FLUX.1-dev and SANA-1.6B.