Text-to-Image Generation on MS-COCO 2017 (val)
19.2591FIDSAMG
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SAMGModel=SD 1.52026.04 | 19.2591 | 0.2055 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CFGModel=SD 1.52026.04 | 19.4704 | 0.2021 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HSOSolver=UniPC, NFE=52025.11 | 19.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DPM-Solver++Diffusion Model=Stable Diffusion, Sampling Steps (T)=202023.12 | 20.51 | 0.782 | — | — | — | — | — | 14,821 | 0.9 | 1.13 | — | — | — | — | — | — | — | |
| I.I.DBase Model=FLUX.1-dev2026.05 | 20.53 | 0.2561 | — | — | — | — | — | — | — | — | — | — | — | 6.43 | — | 0.707 | 0.01 | |
| Faster Diffusion (DDIM + ToMe)Diffusion Model=Stable Diffusion, Sampling Steps (T)=50, Base Sampling Method=DDIM + ToMe2023.12 | 20.73 | 0.781 | — | — | — | — | — | 26,053 | 1.15 | 1.33 | — | — | — | — | — | — | — | |
| Faster Diffusion (DPM-Solver++)Diffusion Model=Stable Diffusion, Sampling Steps (T)=20, Base Sampling Method=DPM-Solver++2023.12 | 20.76 | 0.781 | — | — | — | — | — | 11,743 | 0.46 | 0.64 | — | — | — | — | — | — | — | |
| DPM-Solver++Diffusion Model=DeepFloyd-IF, Sampling Steps (T)=1002023.12 | 20.79 | 0.784 | — | — | — | — | — | 370,525 | 15.19 | 16.09 | — | — | — | — | — | — | — | |
| Faster Diffusion (DPM-Solver++)Diffusion Model=DeepFloyd-IF, Sampling Steps (T)=100, Base Sampling Method=DPM-Solver++2023.12 | 20.85 | 0.785 | — | — | — | — | — | 313,381 | 12.02 | 12.97 | — | — | — | — | — | — | — | |
| SAMGModel=SD3.5-M2026.04 | 20.868 | 0.1941 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Faster DiffusionSampling steps (T)=50, Multi-GPU Parallel=true, Base Method=DDIM2023.12 | 21.08 | 0.783 | — | — | — | — | 560 | — | — | — | — | — | — | — | — | — | — | |
| Faster Diffusion (DDIM)Diffusion Model=Stable Diffusion, Sampling Steps (T)=50, Base Sampling Method=DDIM2023.12 | 21.08 | 0.783 | — | — | — | — | — | 27,350 | 1.21 | 1.42 | — | — | — | — | — | — | — | |
| Faster Diffusion (DPM-Solver)Diffusion Model=Stable Diffusion, Sampling Steps (T)=20, Base Sampling Method=DPM-Solver2023.12 | 21.25 | 0.779 | — | — | — | — | — | 11,743 | 0.46 | 0.64 | — | — | — | — | — | — | — | |
| DPM-SolverDiffusion Model=Stable Diffusion, Sampling Steps (T)=202023.12 | 21.36 | 0.78 | — | — | — | — | — | 14,821 | 0.9 | 1.14 | — | — | — | — | — | — | — | |
| DeepCacheSampling steps (T)=50, Multi-GPU Parallel=false, Base Method=DDIM2023.12 | 21.53 | 0.77 | — | — | — | — | 1,050 | — | — | — | — | — | — | — | — | — | — | |
| CacheMeSampling steps (T)=50, Multi-GPU Parallel=false, Base Method=DDIM2023.12 | 21.62 | 0.775 | — | — | — | — | 1,360 | — | — | — | — | — | — | — | — | — | — | |
| DDIMSampling steps (T)=50, Multi-GPU Parallel=false2023.12 | 21.75 | 0.773 | — | — | — | — | 2,420 | — | — | — | — | — | — | — | — | — | — | |
| DDIMDiffusion Model=Stable Diffusion, Sampling Steps (T)=502023.12 | 21.75 | 0.773 | — | — | — | — | — | 37,050 | 2.23 | 2.42 | — | — | — | — | — | — | — | |
| CFGModel=SD3.5-M2026.04 | 21.9324 | 0.1886 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DDIM + ToMeDiffusion Model=Stable Diffusion, Sampling Steps (T)=502023.12 | 22.32 | 0.782 | — | — | — | — | — | 35,123 | 2.07 | 2.26 | — | — | — | — | — | — | — | |
| DM-NonUniSolver=UniPC, NFE=52025.11 | 23.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HSOSolver=DDIM, NFE=52025.11 | 23.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HSOSolver=UniPC, NFE=42025.11 | 23.26 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Faster Diffusion (DDPM)Diffusion Model=DeepFloyd-IF, Sampling Steps (T)=225, Base Sampling Method=DDPM2023.12 | 23.73 | 0.782 | — | — | — | — | — | 626,523 | 25.61 | 26.27 | — | — | — | — | — | — | — | |
| DDPMDiffusion Model=DeepFloyd-IF, Sampling Steps (T)=2252023.12 | 23.89 | 0.783 | — | — | — | — | — | 734,825 | 33.91 | 34.55 | — | — | — | — | — | — | — | |
| Stable Diffusion v1-5NFE=50, Retrieval=false2026.02 | 24.38 | 0.319 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CADSBase Model=FLUX.1-dev, Variation Level=Low2026.05 | 25.032 | 0.2316 | — | — | — | — | — | — | — | — | — | — | — | 6.563 | — | 0.618 | 0.31 | |
| Stable Diffusion v2-1-baseNFE=50, Retrieval=false2026.02 | 25.33 | 0.33 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAMGModel=SDXL2026.04 | 25.4147 | 0.2014 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ImageRAGTurboNFE=1, Retrieval=true2026.02 | 25.59 | 0.323 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| I.I.D2026.05 | 25.595 | 0.2689 | — | — | — | — | — | — | — | — | — | — | — | 6.227 | — | 0.666 | 0 | |
| Stable Diffusion Turbo v2-1-baseNFE=1, Retrieval=false2026.02 | 26.04 | 0.319 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EDDYBase Model=FLUX.1-dev, Variation Level=Low2026.05 | 26.73 | 0.2538 | — | — | — | — | — | — | — | — | — | — | — | 6.124 | — | 0.604 | 0.41 | |
| AT-EDM-dTFLOPs=4.5, CFG-scale=7.02024.05 | 27.23 | 0.3245 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SD-XL†TFLOPs=6.7, CFG-scale=4.02024.05 | 27.3 | 0.3226 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SDXLConfiguration=Original Model, Devices=12024.06 | 27.43 | 0.3233 | — | 13.81 | 1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| DM-NonUniSolver=UniPC, NFE=42025.11 | 27.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RDMNFE=50, Retrieval=true2026.02 | 27.6 | 0.293 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 2.1, N=2, S=1, Devices=22024.06 | 27.79 | 0.3159 | — | 3.03 | 1.8 | 0.2121 | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SDXL, N=2, S=1, Devices=22024.06 | 27.79 | 0.3221 | — | 8 | 1.7 | 0.2509 | — | — | — | — | — | — | — | — | — | — | — | |
| CADSBase Model=FLUX.1-dev, Variation Level=Medium2026.05 | 27.876 | 0.2253 | — | — | — | — | — | — | — | — | — | — | — | 6.609 | — | 0.575 | 0.57 | |
| SD 2.1Configuration=Original Model, Devices=12024.06 | 27.89 | 0.316 | — | 5.51 | 1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CFGModel=SDXL2026.04 | 27.9759 | 0.1951 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 2.1, N=3, S=1, Devices=32024.06 | 28 | 0.3156 | — | 2.41 | 2.3 | 0.2755 | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SDXL, N=3, S=1, Devices=32024.06 | 28.03 | 0.3205 | — | 5.84 | 2.4 | 0.294 | — | — | — | — | — | — | — | — | — | — | — | |
| AT-EDM-cTFLOPs=4.1, CFG-scale=7.02024.05 | 28.04 | 0.3209 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 2.1, N=4, S=1, Devices=42024.06 | 28.28 | 0.314 | — | 2.1 | 2.6 | 0.3132 | — | — | — | — | — | — | — | — | — | — | — | |
| DyDiTConfiguration=b, MACs(T)=3.74, Latency(s)=6.87, Speed=1.55x, Finetune=true2025.05 | 28.47 | 0.3213 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 2.1, N=2, S=2, Devices=32024.06 | 28.55 | 0.3143 | — | 1.82 | 3 | 0.3458 | — | — | — | — | — | — | — | — | — | — | — | |
| SDTM*Configuration=a, MACs(T)=4.13, Latency(s)=8.02, Speed=1.33x, Finetune=false2025.05 | 28.57 | 0.3249 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SDTMConfiguration=a, MACs(T)=4.20, Latency(s)=8.20, Speed=1.30x, Finetune=false2025.05 | 28.73 | 0.3235 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TokenCacheConfiguration=b, MACs(T)=3.72, Latency(s)=6.97, Speed=1.53x, Finetune=true2025.05 | 28.83 | 0.3208 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SDTM*Configuration=b, MACs(T)=3.62, Latency(s)=6.90, Speed=1.55x, Finetune=false2025.05 | 28.97 | 0.3237 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SDXL, N=2, S=2, Devices=32024.06 | 28.99 | 0.317 | — | 4.91 | 2.8 | 0.3209 | — | — | — | — | — | — | — | — | — | — | — | |
| SD3 MediumMACs(T)=6.01, Latency(s)=10.67, Speed=1.00x, Finetune=false2025.05 | 29.02 | 0.3267 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SDXL, N=4, S=1, Devices=42024.06 | 29.12 | 0.319 | — | 5.12 | 2.7 | 0.3157 | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 2.1, N=3, S=2, Devices=42024.06 | 29.41 | 0.3122 | — | 1.35 | 4 | 0.3778 | — | — | — | — | — | — | — | — | — | — | — | |
| HSOSolver=DDIM, NFE=42025.11 | 29.55 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SDTMConfiguration=b, MACs(T)=3.66, Latency(s)=7.07, Speed=1.51x, Finetune=false2025.05 | 29.6 | 0.3224 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AT-EDM-bTFLOPs=3.6, CFG-scale=7.02024.05 | 29.8 | 0.3095 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 1.5, N=3, S=1, Devices=32024.06 | 29.87 | 0.3058 | — | 1.23 | 2.2 | 0.2645 | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 1.5, N=2, S=1, Devices=22024.06 | 29.94 | 0.3062 | — | 1.52 | 1.8 | 0.1988 | — | — | — | — | — | — | — | — | — | — | — | |
| SD 1.5Configuration=Original Model, Devices=12024.06 | 29.96 | 0.3063 | — | 2.7 | 1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AsyncDiffBase Model=SD 1.5, N=4, S=1, Devices=42024.06 | 30.1 | 0.3052 | — | 1.01 | 2.6 | 0.3073 | — | — | — | — | — | — | — | — | — | — | — | |
| DM-NonUniSolver=DDIM, NFE=52025.11 | 30.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PGGuidance Level=Low2026.05 | 30.131 | 0.2627 | — | — | — | — | — | — | — | — | — | — | — | 5.898 | — | 0.611 | 0.0002 | |
| AsyncDiffBase Model=SDXL, N=3, S=2, Devices=42024.06 | 30.27 | 0.314 | — | 3.65 | 3.8 | 0.3556 | — | — | — | — | — | — | — | — | — | — | — | |
| Lumina-mGPT (7B)Decoding Strategy=Autoregressive (Standard)2025.12 | 30.77 | 0.3152 | — | 87.23 | — | — | — | — | — | — | 1 | 1 | 0.2857 | — | — | — | — | |
| PGBase Model=FLUX.1-dev, Variation Level=Low2026.05 | 30.822 | 0.2493 | — | — | — | — | — | — | — | — | — | — | — | 6.013 | — | 0.627 | 2.11 | |
| AsyncDiffBase Model=SD 1.5, N=3, S=2, Devices=42024.06 | 30.89 | 0.3017 | — | 0.72 | 3.7 | 0.3811 | — | — | — | — | — | — | — | — | — | — | — | |
| Lumina-mGPT (7B)Decoding Strategy=Speculative Jacobi Decoding (SJD)2025.12 | 30.9 | 0.314 | — | 42.73 | — | — | — | — | — | — | 2.04 | 2.22 | 0.2838 | — | — | — | — | |
| AsyncDiffBase Model=SD 1.5, N=2, S=2, Devices=32024.06 | 30.98 | 0.3046 | — | 0.94 | 2.9 | 0.3232 | — | — | — | — | — | — | — | — | — | — | — | |
| CADSBase Model=FLUX.1-dev, Variation Level=High2026.05 | 31.42 | 0.2158 | — | — | — | — | — | — | — | — | — | — | — | 6.615 | — | 0.522 | 0.8 | |
| Lumina-mGPT (7B)Decoding Strategy=SJD with token reuse (SJD++)2025.12 | 31.48 | 0.3127 | — | 27.92 | — | — | — | — | — | — | 3.12 | 6.44 | 0.2817 | — | — | — | — | |
| EDDYGuidance Level=Low2026.05 | 31.507 | 0.2652 | — | — | — | — | — | — | — | — | — | — | — | 6.234 | — | 0.622 | 0 | |
| SD-XLTFLOPs=6.7, CFG-scale=7.02024.05 | 31.94 | 0.3284 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LlamaGen (0.8B)Decoding Strategy=Autoregressive (Standard)2025.12 | 32.25 | 0.3045 | — | 46.02 | — | — | — | — | — | — | 1 | 1 | 0.2674 | — | — | — | — | |
| LlamaGen (0.8B)Decoding Strategy=Speculative Jacobi Decoding (SJD)2025.12 | 32.4 | 0.3033 | — | 33.18 | — | — | — | — | — | — | 1.39 | 2.1 | 0.2651 | — | — | — | — | |
| ToMe-dTFLOPs=4.6, CFG-scale=7.02024.05 | 32.46 | 0.3235 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LlamaGen (0.8B)Decoding Strategy=SJD with token reuse (SJD++)2025.12 | 32.58 | 0.302 | — | 19.84 | — | — | — | — | — | — | 2.32 | 3.26 | 0.263 | — | — | — | — | |
| CDGBackbone=SD3.52026.03 | 33.07 | 0.3196 | — | — | — | — | — | — | — | — | — | — | — | 6.26 | 92.61 | — | — | |
| Janus-Pro (1B)Decoding Strategy=Autoregressive (Standard)2025.12 | 33.97 | 0.3149 | — | 14.5 | — | — | — | — | — | — | 1 | 1 | 0.2838 | — | — | — | — | |
| CDGBackbone=SD32026.03 | 34.05 | 0.32 | — | — | — | — | — | — | — | — | — | — | — | 5.7 | 92.4 | — | — | |
| Janus-Pro (1B)Decoding Strategy=Speculative Jacobi Decoding (SJD)2025.12 | 34.2 | 0.3138 | — | 8.01 | — | — | — | — | — | — | 1.81 | 1.96 | 0.2819 | — | — | — | — | |
| Janus-Pro (1B)Decoding Strategy=SJD with token reuse (SJD++)2025.12 | 34.4 | 0.3127 | — | 6.29 | — | — | — | — | — | — | 2.31 | 2.47 | 0.2803 | — | — | — | — | |
| EDDYBase Model=FLUX.1-dev, Variation Level=Medium2026.05 | 34.512 | 0.2498 | — | — | — | — | — | — | — | — | — | — | — | 5.88 | — | 0.537 | 1.23 | |
| CFGBackbone=SD3.52026.03 | 34.56 | 0.3185 | — | — | — | — | — | — | — | — | — | — | — | 6.21 | 91.94 | — | — | |
| CADSBackbone=SD3.52026.03 | 34.58 | 0.3186 | — | — | — | — | — | — | — | — | — | — | — | 6.21 | 91.83 | — | — | |
| DNPBackbone=SD32026.03 | 34.68 | 0.313 | — | — | — | — | — | — | — | — | — | — | — | 5.51 | 89.65 | — | — | |
| AT-EDMConfiguration=a, MACs(T)=4.23, Latency(s)=8.14, Speed=1.31x, Finetune=false2025.05 | 34.72 | 0.3195 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ToMe-cTFLOPs=4.1, CFG-scale=7.02024.05 | 35.27 | 0.3198 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ICGBackbone=SD3.52026.03 | 35.41 | 0.317 | — | — | — | — | — | — | — | — | — | — | — | 6.32 | 91.8 | — | — | |
| Emu3-Gen (8B)Decoding Strategy=Autoregressive (Standard)2025.12 | 35.46 | 0.2996 | — | 375.29 | — | — | — | — | — | — | 1 | 1 | 0.2563 | — | — | — | — | |
| Emu3-Gen (8B)Decoding Strategy=Speculative Jacobi Decoding (SJD)2025.12 | 35.6 | 0.2995 | — | 207.61 | — | — | — | — | — | — | 1.81 | 2.01 | 0.2542 | — | — | — | — | |
| CFGBackbone=SD32026.03 | 35.69 | 0.3173 | — | — | — | — | — | — | — | — | — | — | — | 5.66 | 91.44 | — | — | |
| Emu3-Gen (8B)Decoding Strategy=SJD with token reuse (SJD++)2025.12 | 35.79 | 0.2983 | — | 124.62 | — | — | — | — | — | — | 3.01 | 7.5 | 0.2524 | — | — | — | — | |
| SDITSteps=14, Ratio=0.5, time (s)=6.892026.01 | 35.85 | 0.2151 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CADSBackbone=SD32026.03 | 36.16 | 0.3172 | — | — | — | — | — | — | — | — | — | — | — | 5.65 | 91.44 | — | — | |
| Latent Consistency ModelNFE=4, Retrieval=false2026.02 | 36.52 | 0.307 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SimpleAR (0.5B)Decoding Strategy=Autoregressive (Standard)2025.12 | 36.78 | 0.2832 | — | 83.06 | — | — | — | — | — | — | 1 | 1 | 0.2393 | — | — | — | — |