LaCon: Late-Constraint Diffusion for Steerable Guided Image Synthesis

Benchmark Model Rank Results
conditional-text-to-image-synthesis-on-cocoLCDG#1FID: 20.27
conditional-text-to-image-synthesis-on-cocoLCDG (Color, evaluated under image palette)#2FID: 20.61CLIP Score: 0.2580
conditional-text-to-image-synthesis-on-cocoLCDG (Mask)#3FID: 20.94CLIP Score: 0.2617
conditional-text-to-image-synthesis-on-cocoLCDG (Edge)#4FID: 21.02
conditional-text-to-image-synthesis-on-cocoT2I-Adapter (Sketch)#5FID: 21.72CLIP Score: 0.2580
conditional-text-to-image-synthesis-on-cocoT2I-Adapter (Color, evaluated under image palette)#6FID: 26.54CLIP Score: 0.2613
conditional-text-to-image-synthesis-on-cocoSD (text)#7FID: 27.99CLIP Score: 0.2673
conditional-text-to-image-synthesis-on-cocoControlNet (HED Edge)#8FID: 28.09CLIP Score: 0.2525
conditional-text-to-image-synthesis-on-cocoT2I-Adapter (Color, evaluated under color stroke)#9FID: 30.84
conditional-text-to-image-synthesis-on-cocoSD using SDEdit (evaluated under color stroke)#10FID: 32.93CLIP Score: 0.2257
conditional-text-to-image-synthesis-on-cocoSD using SDEdit#11FID: 71.16
conditional-text-to-image-synthesis-on-cocoSD using SDEdit (evaluated under image palette)#12CLIP Score: 0.2138