Linum (auto-discovered)

A research team reports that its JiT-DDT encoder-decoder architecture trains a text-to-image model with 3.6× fewer GPU-hours than its Linum v2 baseline while generating images with 4× the pixels. The …

BentoML's engineering team reports that most gains in image and video model performance since Stable Diffusion 3 come from data filtering, rebalancing, annotation, and synthetic data generation, not f…

A blog post from an unnamed AI research team details how data filtering for generative video pre-training has evolved since 2024, emphasizing that most model improvements come from data filtering, reb…