ENHANCING TEXT-TO-IMAGE ALIGNMENT FOR LONG TEXTS

Summary from the TLDR AI newsletter

LongAlign improves text-to-image (T2I) diffusion models' ability to handle long text inputs by introducing segment-level encoding and decomposed preference optimization.