Text-to-image generation on COCO 2017 (val)
11FIDBADiff
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BADiffBitrate Regime=High, Bitrate Range (bpp)=1.0–2.02025.10 | 11 | — | — | — | — | — | — | |
| Cascade (SD + LIC)Bitrate Regime=High, Bitrate Range (bpp)=1.0–2.02025.10 | 13.1 | — | — | — | — | — | — | |
| Res-TuningParam. (M)=2.54, Mem. (GB)=54.49, Train (Hour/Epoch)=1.382023.10 | 13.96 | — | — | — | — | — | — | |
| SD v1.5 + LoRAParam. (M)=9.96, Mem. (GB)=61.03, Train (Hour/Epoch)=1.422023.10 | 14.5 | — | — | — | — | — | — | |
| SD v1.5 + AdapterParam. (M)=2.51, Mem. (GB)=54.30, Train (Hour/Epoch)=1.302023.10 | 14.73 | — | — | — | — | — | — | |
| Cascade (SD + BPG)Bitrate Regime=High, Bitrate Range (bpp)=1.0–2.02025.10 | 14.8 | — | — | — | — | — | — | |
| SD v1.5 + Full TuningParam. (M)=862, Mem. (GB)=72.77, Train (Hour/Epoch)=1.982023.10 | 14.85 | — | — | — | — | — | — | |
| Res-Tuning BypassParam. (M)=3.76, Mem. (GB)=21.35, Train (Hour/Epoch)=0.822023.10 | 14.89 | — | — | — | — | — | — | |
| SD v1.5 + PromptParam. (M)=1.25, Mem. (GB)=63.70, Train (Hour/Epoch)=2.172023.10 | 14.9 | — | — | — | — | — | — | |
| SD v1.5 + PrefixParam. (M)=4.99, Mem. (GB)=64.91, Train (Hour/Epoch)=2.202023.10 | 15.36 | — | — | — | — | — | — | |
| SD v1.5tuning_mode=Zero-shot2023.10 | 15.48 | — | — | — | — | — | — | |
| BADiffBitrate Regime=Medium, Bitrate Range (bpp)=0.5–1.02025.10 | 16.2 | — | — | — | — | — | — | |
| DPM-Solver (20) + LICBitrate Regime=High, Bitrate Range (bpp)=1.0–2.02025.10 | 16.3 | — | — | — | — | — | — | |
| Early-Stop + LICBitrate Regime=High, Bitrate Range (bpp)=1.0–2.02025.10 | 18 | — | — | — | — | — | — | |
| Cascade (SD + LIC)Bitrate Regime=Medium, Bitrate Range (bpp)=0.5–1.02025.10 | 19.2 | — | — | — | — | — | — | |
| Cascade (SD + BPG)Bitrate Regime=Medium, Bitrate Range (bpp)=0.5–1.02025.10 | 21.4 | — | — | — | — | — | — | |
| DMD2EMPURPLE=Yes, Cosine similarity=Yes, Resolution=1024 x 10242026.07 | 22.24 | 30.81 | — | — | — | — | — | |
| DMD2EMPURPLE=Yes, Cosine similarity=No, Resolution=1024 x 10242026.07 | 22.57 | 30.81 | — | — | — | — | — | |
| SDXL-LightningEMPURPLE=Yes, Cosine similarity=Yes, Resolution=1024 x 10242026.07 | 24.46 | 29.53 | — | — | — | — | — | |
| SDXL-LightningEMPURPLE=Yes, Cosine similarity=No, Resolution=1024 x 10242026.07 | 24.72 | 29.46 | — | — | — | — | — | |
| DPM-Solver (20) + LICBitrate Regime=Medium, Bitrate Range (bpp)=0.5–1.02025.10 | 25.1 | — | — | — | — | — | — | |
| DMD2EMPURPLE=No, Cosine similarity=–, Resolution=1024 x 10242026.07 | 25.36 | 31.77 | — | — | — | — | — | |
| Flash SDXLEMPURPLE=Yes, Cosine similarity=Yes, Resolution=1024 x 10242026.07 | 25.56 | 29.49 | — | — | — | — | — | |
| Flash SDXLEMPURPLE=Yes, Cosine similarity=No, Resolution=1024 x 10242026.07 | 25.62 | 29.49 | — | — | — | — | — | |
| GEAR-1.0BCFG=w/2026.06 | 25.74 | 31.39 | — | — | — | — | — | |
| Rectified DiffusionSteps=42026.06 | 25.81 | — | — | — | — | — | — | |
| MA-ReflowSteps=4, K=42026.06 | 25.94 | — | — | — | — | — | — | |
| SDv3.5 w/ VAGSCFG=7, κ=1.02026.05 | 26.07 | 32.84 | 35.15 | 8.28 | 29.53 | 35.21 | 71.62 | |
| BADiffBitrate Regime=Low, Bitrate Range (bpp)=0.2–0.52025.10 | 26.1 | — | — | — | — | — | — | |
| DMD2Steps=42026.06 | 26.19 | — | — | — | — | — | — | |
| PeRFlowSteps=42026.06 | 27.06 | — | — | — | — | — | — | |
| LCMSteps=42026.06 | 27.09 | — | — | — | — | — | — | |
| Flash SDXLEMPURPLE=No, Cosine similarity=–, Resolution=1024 x 10242026.07 | 27.09 | 31.24 | — | — | — | — | — | |
| Hyper SDXLEMPURPLE=Yes, Cosine similarity=No, Resolution=1024 x 10242026.07 | 27.1 | 30.06 | — | — | — | — | — | |
| Early-Stop + LICBitrate Regime=Medium, Bitrate Range (bpp)=0.5–1.02025.10 | 27.5 | — | — | — | — | — | — | |
| SDv3.5 w/ A-Euler2026.05 | 27.64 | 32.84 | 34.63 | 7.96 | 29.09 | 35.17 | 71.23 | |
| SDXL-turboEMPURPLE=Yes, Cosine similarity=No, Resolution=512 x 5122026.07 | 27.69 | 31.52 | — | — | — | — | — | |
| LlamaGen-REPA-1.0BCFG=w/2026.06 | 27.99 | 30.8 | — | — | — | — | — | |
| Hyper SDXLEMPURPLE=Yes, Cosine similarity=Yes, Resolution=1024 x 10242026.07 | 28.12 | 30.08 | — | — | — | — | — | |
| SDv3.5CFG=72026.05 | 28.46 | 32.84 | 33.64 | 7.99 | 29.17 | 35.11 | 71.45 | |
| SDXL-LightningEMPURPLE=No, Cosine similarity=–, Resolution=1024 x 10242026.07 | 28.99 | 31.21 | — | — | — | — | — | |
| GEAR-1.0BCFG=w/o2026.06 | 29.66 | 27.29 | — | — | — | — | — | |
| LlamaGen-REPA-1.0BCFG=w/o2026.06 | 30.04 | 27.12 | — | — | — | — | — | |
| SDXL-LightningSteps=42026.06 | 30.16 | — | — | — | — | — | — | |
| HyperSDSteps=42026.06 | 30.58 | — | — | — | — | — | — | |
| Cascade (SD + LIC)Bitrate Regime=Low, Bitrate Range (bpp)=0.2–0.52025.10 | 30.7 | — | — | — | — | — | — | |
| SDXL-TurboSteps=42026.06 | 30.92 | — | — | — | — | — | — | |
| Hyper SDXLEMPURPLE=No, Cosine similarity=–, Resolution=1024 x 10242026.07 | 31.79 | 31.46 | — | — | — | — | — | |
| SDXL-turboEMPURPLE=No, Cosine similarity=–, Resolution=512 x 5122026.07 | 32.08 | 31.67 | — | — | — | — | — | |
| Cascade (SD + BPG)Bitrate Regime=Low, Bitrate Range (bpp)=0.2–0.52025.10 | 33.5 | — | — | — | — | — | — | |
| DPM-Solver (20) + LICBitrate Regime=Low, Bitrate Range (bpp)=0.2–0.52025.10 | 36.5 | — | — | — | — | — | — | |
| SDv3.5 w/ Self-Guidance2026.05 | 40.98 | 31.21 | 28.54 | 6.52 | 26.18 | 31.59 | 62.26 | |
| Early-Stop + LICBitrate Regime=Low, Bitrate Range (bpp)=0.2–0.52025.10 | 41.8 | — | — | — | — | — | — | |
| BK-SDMVariant=Base, MACs=223.81G, Throughput=0.344, Speed=1.45x2023.12 | — | 29.47 | — | — | — | — | — | |
| BK-SDMVariant=Small, MACs=217.78G, Throughput=0.397, Speed=1.68x2023.12 | — | 27.96 | — | — | — | — | — | |
| BK-SDMVariant=Tiny, MACs=205.09G, Throughput=0.415, Speed=1.76x2023.12 | — | 27.41 | — | — | — | — | — | |
| DeepCacheN=5, MACs=130.45G, Throughput=0.5, Speed=2.11x2023.12 | — | 30.23 | — | — | — | — | — | |
| PLMSSampling Steps=50, MACs=338.83G, Throughput=0.237, Speed=1.00x2023.12 | — | 30.3 | — | — | — | — | — | |
| PLMSSampling Steps=25, MACs=169.42G, Throughput=0.453, Speed=1.91x2023.12 | — | 29.99 | — | — | — | — | — |