Text-to-image generation on DrawBench (Efficiency and Fidelity Metrics)
5.5Latency (s)X-Slim(C2F)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| X-Slim(C2F)setting=fast2025.12 | 5.5 | 4.97 | 0.9806 | 0.3153 | 19.241 | 0.7321 | 0.2919 | |
| SpectrumModel=FLUX.1 [17], alpha (α)=3.02026.03 | 5.6 | 4.79 | 1 | 27.64 | 22.21 | 0.788 | 0.261 | |
| 20% stepsSteps=102025.12 | 5.74 | 4.75 | 0.7878 | 0.3139 | 14.529 | 0.6153 | 0.4733 | |
| Uniform CacheN=52025.12 | 5.8 | 4.71 | 0.7991 | 0.3142 | 14.53 | 0.612 | 0.4755 | |
| SpectrumModel=Stable Diffusion 3.5-Large [4], alpha (α)=3.02026.03 | 5.81 | 4.32 | 0.82 | 28 | 15.68 | 0.62 | 0.43 | |
| X-Slim(S)setting=fast2025.12 | 5.83 | 4.69 | 0.981 | 0.3156 | 19.56 | 0.752 | 0.2884 | |
| TeaCacheδ=12025.12 | 5.96 | 4.58 | 0.8619 | 0.3137 | 15.472 | 0.6437 | 0.4378 | |
| TaylorSeerModel=FLUX.1 [17], N=6, O=12026.03 | 6.48 | 4.14 | 1 | 27.6 | 20.24 | 0.785 | 0.294 | |
| FORAModel=FLUX.1 [17], N=62026.03 | 6.49 | 4.13 | 0.85 | 27.3 | 14.33 | 0.613 | 0.505 | |
| TaylorSeerModel=FLUX.1 [17], N=6, O=22026.03 | 6.52 | 3.99 | 0.99 | 27.39 | 17.41 | 0.708 | 0.389 | |
| FORAModel=Stable Diffusion 3.5-Large [4], N=62026.03 | 6.73 | 3.72 | 0.02 | 26.19 | 9.05 | 0.389 | 0.687 | |
| TeaCacheModel=FLUX.1 [17], delta (δ)=0.82026.03 | 6.75 | 3.85 | 1 | 27.63 | 16.71 | 0.695 | 0.406 | |
| TaylorSeerModel=Stable Diffusion 3.5-Large [4], N=6, O=22026.03 | 6.76 | 3.71 | 0.1 | 25.9 | 9.36 | 0.429 | 0.636 | |
| TaylorSeerModel=Stable Diffusion 3.5-Large [4], N=6, O=12026.03 | 6.78 | 3.69 | 0.32 | 27.27 | 11.18 | 0.505 | 0.535 | |
| TeaCacheModel=Stable Diffusion 3.5-Large [4], delta (δ)=0.82026.03 | 7.17 | 3.49 | 0.39 | 27.08 | 10.04 | 0.472 | 0.589 | |
| ToCaN=82025.12 | 7.19 | 3.8 | 0.8608 | 0.3039 | 16.351 | 0.5243 | 0.5319 | |
| Δ-DitN=52025.12 | 7.51 | 3.64 | 0.8234 | 0.3019 | 14.103 | 0.6391 | 0.4588 | |
| TaylorSeerN=62025.12 | 7.59 | 3.6 | 0.9688 | 0.3182 | 15.78 | 0.6514 | 0.4014 | |
| 15 stepsModel=Stable Diffusion 3.5-Large [4], Sampling Steps=152026.03 | 7.71 | 3.25 | 0.85 | 28.75 | 10.51 | 0.453 | 0.595 | |
| SpectrumModel=FLUX.1 [17], alpha (α)=0.752026.03 | 7.73 | 3.47 | 0.99 | 27.65 | 24.32 | 0.854 | 0.217 | |
| SpectrumModel=Stable Diffusion 3.5-Large [4], alpha (α)=0.752026.03 | 7.82 | 3.21 | 0.97 | 28.61 | 17.83 | 0.743 | 0.305 | |
| 15 stepsModel=FLUX.1 [17], Sampling Steps=152026.03 | 8.04 | 3.24 | 1 | 27.56 | 15.77 | 0.673 | 0.432 | |
| X-Slim(C2F)setting=slow2025.12 | 8.11 | 3.37 | 0.9902 | 0.3162 | 21.3 | 0.8097 | 0.2105 | |
| 30% stepsSteps=152025.12 | 8.4 | 3.25 | 0.9016 | 0.3153 | 15.494 | 0.6585 | 0.4035 | |
| FORAModel=FLUX.1 [17], N=42026.03 | 8.4 | 3.19 | 0.97 | 27.55 | 15.15 | 0.651 | 0.454 | |
| TeaCacheδ=0.62025.12 | 8.42 | 3.24 | 0.9718 | 0.3145 | 16.938 | 0.7006 | 0.3433 | |
| TaylorSeerModel=FLUX.1 [17], N=4, O=12026.03 | 8.55 | 3.13 | 0.99 | 27.6 | 22.31 | 0.841 | 0.215 | |
| TaylorSeerModel=FLUX.1 [17], N=4, O=22026.03 | 8.59 | 3.03 | 1.02 | 27.61 | 20.76 | 0.812 | 0.247 | |
| FORAModel=Stable Diffusion 3.5-Large [4], N=42026.03 | 8.63 | 2.9 | 0.27 | 27.03 | 9.54 | 0.437 | 0.606 | |
| TaylorSeerModel=Stable Diffusion 3.5-Large [4], N=4, O=22026.03 | 8.7 | 2.88 | 0.44 | 27.24 | 11.12 | 0.513 | 0.529 | |
| ToCaN=52025.12 | 8.73 | 3.13 | 0.8951 | 0.3105 | 17.288 | 0.6007 | 0.4217 | |
| TaylorSeerModel=Stable Diffusion 3.5-Large [4], N=4, O=12026.03 | 8.75 | 2.86 | 0.82 | 27.91 | 13.13 | 0.605 | 0.428 | |
| X-Slim(S)setting=slow2025.12 | 9.08 | 3.01 | 0.9901 | 0.3164 | 21.717 | 0.8162 | 0.1905 | |
| Uniform CacheN=32025.12 | 9.58 | 2.85 | 0.8959 | 0.315 | 15.816 | 0.6726 | 0.3828 | |
| TeaCacheModel=FLUX.1 [17], delta (δ)=0.42026.03 | 10.66 | 2.44 | 1.03 | 27.68 | 18.7 | 0.762 | 0.311 | |
| TeaCacheModel=Stable Diffusion 3.5-Large [4], delta (δ)=0.42026.03 | 10.75 | 2.33 | 0.78 | 27.46 | 11.89 | 0.527 | 0.512 | |
| TaylorSeerN=32025.12 | 11.63 | 2.35 | 0.9814 | 0.3158 | 19.603 | 0.782 | 0.2231 | |
| Δ-DitN=32025.12 | 12.36 | 2.21 | 0.8987 | 0.3151 | 15.633 | 0.6649 | 0.3915 | |
| 25 stepsModel=Stable Diffusion 3.5-Large [4], Sampling Steps=252026.03 | 12.67 | 1.98 | 1.02 | 28.78 | 12.41 | 0.593 | 0.464 | |
| 25 stepsModel=FLUX.1 [17], Sampling Steps=252026.03 | 13.23 | 1.97 | 1.01 | 27.58 | 18.11 | 0.752 | 0.322 | |
| ToCaModel=FLUX.1 [17], N=6, R=0.92026.03 | 13.38 | 1.95 | 0.99 | 27.61 | 16.49 | 0.666 | 0.439 | |
| 50% stepsSteps=252025.12 | 13.78 | 1.98 | 0.964 | 0.3152 | 17.922 | 0.7399 | 0.2867 | |
| ToCaModel=Stable Diffusion 3.5-Large [4], N=6, R=0.92026.03 | 13.99 | 1.79 | 0.17 | 26.33 | 9.53 | 0.446 | 0.611 | |
| ToCaModel=FLUX.1 [17], N=4, R=0.92026.03 | 16.93 | 1.58 | 1.03 | 27.63 | 17.18 | 0.724 | 0.362 | |
| ToCaModel=Stable Diffusion 3.5-Large [4], N=4, R=0.92026.03 | 17.27 | 1.45 | 0.51 | 27.19 | 10.55 | 0.483 | 0.557 | |
| 50 stepsModel=Stable Diffusion 3.5-Large [4], Sampling Steps=502026.03 | 25.05 | 1 | 1.05 | 28.66 | — | — | — | |
| 50 stepsModel=FLUX.1 [17], Sampling Steps=502026.03 | 26.03 | 1 | 1 | 27.49 | — | — | — | |
| OriginalSteps=502025.12 | 27.32 | 1 | 0.9886 | 0.3159 | — | — | — |