ResearchBenchmarksText-to-Image Generation on T2ICompBench (Complex Category)Text-to-Image Generation on T2ICompBench (Complex Category)Follow1.38ImageReward (A-A)CO30.898481.023491.14851.27351Sep 30, 2025Evaluation resultsMethodMethodLinksImageReward (A-A)ImageReward (A-O)ImageReward (O-O)BLIP-VQA Accuracy (A-A)BLIP-VQA Accuracy (A-O)BLIP-VQA Accuracy (O-O)CO3backbone=PixArt-Σbackbone=PixArt-Σ2025.091.381.771.08174.686.458.1PixArt-Σbackbone=PixArt-Σbackbone=PixArt-Σ2025.091.0661.6971.02369.186.155.9Comp-diffbackbone=PixArt-Σbackbone=PixArt-Σ2025.090.9170.0470.33920.859.147.2