Text-to-image generation on MJHQ-30K 1024x1024
5.99FIDSANA-1.5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SANA-1.5AutoEncoders=DC-AE(f32c32p1), Token-nums=32×32, Params (B)=4.8, Throughput (img/s)=0.262026.03 | 5.99 | 29.23 | |
| PixArt-ΣAutoEncoders=—, Token-nums=64×64, Params (B)=0.6, Throughput (img/s)=0.402026.03 | 6.15 | 28.26 | |
| Hunyuan-DiTAutoEncoders=—, Token-nums=64×64, Params (B)=1.5, Throughput (img/s)=0.052026.03 | 6.54 | 28.19 | |
| FLUX-devAutoEncoders=FLUX-VAE(f8c16p2), Token-nums=64×64, Params (B)=12, Throughput (img/s)=0.042026.03 | 10.15 | 27.47 | |
| SD3.5-mediumAutoEncoders=SD3-VAE (f8c16p2), Token-nums=64×64, Params (B)=2.5, Throughput (img/s)=0.252026.03 | 10.31 | 29.74 | |
| DA-VAE (SD3.5-M)AutoEncoders=DA-VAE (f16c32p2), Token-nums=32×32, Params (B)=2.5, Throughput (img/s)=1.032026.03 | 10.91 | 31.91 | |
| SD3-mediumAutoEncoders=SD3-VAE (f8c16p2), Token-nums=64×64, Params (B)=2.0, Throughput (img/s)=0.362026.03 | 11.92 | 27.83 | |
| SD3.5-medium†AutoEncoders=SD3-VAE (f8c16p2), Token-nums=32×32, Params (B)=2.5, Throughput (img/s)=1.032026.03 | 12.04 | 30.17 |