notesum.ai

Published at October 24

Diff-Instruct++: Training One-step Text-to-image Generator Model to Align with Human Preferences

cs.CL

cs.AI

Released Date: October 24, 2024

Authors: Weijian Luo¹

Aff.: ¹Peking University

Arxiv: https://arxiv.org/abs/2410.18881v1

Model	Animation	Concept-art	Painting	Photo	Average
GLIDE Nichol et al. (2021)	$23.34$	$23.08$	$23.27$	$24.50$	$23.55$
LAFITE Zhou et al. (2022)	$24.63$	$24.38$	$24.43$	$25.81$	$24.81$
VQ-Diffusion Gu et al. (2022)	$24.97$	$24.70$	$25.01$	$25.71$	$25.10$
FuseDream Liu et al. (2021)	$25.26$	$25.15$	$25.13$	$25.57$	$25.28$
Latent Diffusion Rombach et al. (2022)	$25.73$	$25.15$	$25.25$	$26.97$	$25.78$
CogView2 Ding et al. (2022)	$26.50$	$26.59$	$26.33$	$26.44$	$26.47$
DALL·E mini	$26.10$	$25.56$	$25.56$	$26.12$	$25.83$
Versatile Diffusion Xu et al. (2023b)	$26.59$	$26.28$	$26.43$	$27.05$	$26.59$
VQGAN + CLIP Esser et al. (2021)	$26.44$	$26.53$	$26.47$	$26.12$	$26.39$
DALL·E 2 Ramesh et al. (2022)	$27.34$	$26.54$	$26.68$	$27.24$	$26.95$
Stable Diffusion v1.4 Rombach et al. (2022)	$27.26$	$26.61$	$26.66$	$27.27$	$26.95$
Stable Diffusion v2.0 Rombach et al. (2022)	$27.48$	$26.89$	$26.86$	$27.46$	$27.17$
Epic Diffusion	$27.57$	$26.96$	$27.03$	$27.49$	$27.26$
DeepFloyd-XL	$27.64$	$26.83$	$26.86$	$27.75$	$27.27$
Openjourney	$27.85$	$27.18$	$27.25$	$27.53$	$27.45$
MajicMix Realistic	$27.88$	$27.19$	$27.22$	$27.64$	$27.48$
ChilloutMix	$27.92$	$27.29$	$27.32$	$27.61$	$27.54$
Deliberate	$28.13$	$27.46$	$27.45$	$27.62$	$27.67$
Realistic Vision	$28.22$	$27.53$	$27.56$	$27.75$	$27.77$
SDXL-base(Podell et al., 2023)	$28.42$	$27.63$	$27.60$	$27.29$	$27.73$
SDXL-Refiner(Podell et al., 2023)	$28.45$	$27.66$	$27.67$	$27.46$	$27.80$
Dreamlike Photoreal 2.0	$28.24$	$27.60$	$27.59$	$27.99$	$27.86$
SD15-15Step(Rombach et al., 2022)	$26.76$	$26.37$	$26.41$	$27.12$	$26.66$
SD15-25Step(Rombach et al., 2022)	$27.04$	$26.57$	$26.61$	$27.30$	$26.88$
SD15-DPO-15Step(Wallace et al., 2024)	$27.11$	$26.75$	$26.70$	$27.30$	$26.97$
SD15-DPO-25Step(Wallace et al., 2024)	$27.54$	$26.97$	$26.99$	$27.49$	$27.25$
SD15-LCM-1Step(Luo et al., 2023a)	$23.35$	$23.41$	$23.53$	$23.81$	$23.52$
SD15-LCM-4Step(Luo et al., 2023a)	$26.42$	$25.79$	$25.95$	$26.91$	$26.27$
SD15-TCD-1Step(Zheng et al., 2024)	$23.37$	$23.16$	$23.26$	$23.88$	$23.42$
SD15-TCD-4Step(Zheng et al., 2024)	$26.67$	$26.25$	$26.26$	$27.19$	$26.59$
SD15-Hyper-1Step(Ren et al., 2024)	$27.76$	$27.36$	$27.41$	$27.63$	$27.54$
SD15-Hyper-4Step(Ren et al., 2024)	$28.04$	$27.39$	$27.42$	$27.89$	$27.69$
SD15-Instaflow-1Step(Liu et al., 2023)	$26.07$	$25.80$	$25.89$	$26.32$	$26.02$
SD15-PeReflow-1Step(Yan et al., 2024)	$25.70$	$25.45$	$25.57$	$25.96$	$25.67$
SD15-BOOT-1Step(Gu et al., 2023)	$25.29$	$24.40$	$24.61$	$25.16$	$24.86$
SD21-SwiftBrush-1Step(Nguyen & Tran, 2023)	$26.91$	$26.32$	$26.37$	$27.21$	$26.70$
SD15-SiDLSG-1Step(Zhou et al., 2024a)	$27.39$	$26.65$	$26.58$	$27.30$	$26.98$
SD21-SiDLSG-1Step(Zhou et al., 2024a)	$27.42$	$26.81$	$26.79$	$27.31$	$27.08$
SD21-TURBO-1Step(Sauer et al., 2023b)	$27.48$	$26.86$	$27.46$	$26.89$	$27.71$
SDXL-DMD2-1Step-1024(Yin et al., 2024)	$27.67$	$27.02$	$27.01$	$26.94$	$27.16$
SDXL-DMD2-4Step-1024(Yin et al., 2024)	28.97	$27.99$	$27.90$	$28.28$	$28.29$
SDXL-DMD2-1Step-512(Yin et al., 2024)	$27.70$	$27.07$	$27.02$	$26.94$	$27.18$
SDXL-DMD2-4Step-512(Yin et al., 2024)	$27.22$	$26.65$	$26.62$	$26.57$	$26.76$
SD15-DMD2-1Step-512(Yin et al., 2024)	$26.31$	$25.75$	$25.78$	$26.59$	$26.11$
PixelArt- $\alpha$ -25Step-512(Chen et al., 2023)	$28.77$	$27.92$	$27.96$	$28.37$	$28.25$
PixelArt- $\alpha$ -15Step-512(Chen et al., 2023)	$28.68$	$27.85$	$27.87$	$28.29$	$28.17$
SD15-DI++-1Step( $\alpha_{r}=100,\alpha_{c}=1.5$ )	$28.42$	$27.84$	$28.01$	$28.19$	$28.12$
DiT-DI++-1Step( $\alpha_{r}=10,\alpha_{c}=4.5$ )	$28.91$	28.25	28.28	28.50	28.48