Prompt-to-prompt semantic composition on GenAI-Bench
46.8CLIP-T (S*T)Single Prompt
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Single PromptModel=SANA1.5, Mask=false, Layers / Steps=none2026.05 | 46.8 | 70 | 66.8 | 55.6 | 76 | 73.2 | 18 | 46.6 | 38.6 | |
| Single PromptModel=FLUX.1-schnell, Mask=false, Layers / Steps=none2026.05 | 46.5 | 70.3 | 66.3 | 52.7 | 74.5 | 70.8 | 15.5 | 45.4 | 34.2 | |
| masked activation-transport mechanismModel=Qwen-Image, Mask=true, Layers / Steps=mid2026.05 | 44.5 | 67 | 66.4 | 55.4 | 76 | 72.9 | 18 | 48.6 | 37.1 | |
| Single PromptModel=FLUX.1-dev, Mask=false, Layers / Steps=none2026.05 | 44.2 | 68.7 | 64.4 | 54.9 | 76 | 72.2 | 14.4 | 43.9 | 32.9 | |
| masked activation-transport mechanismModel=FLUX.1-schnell, Mask=true, Layers / Steps=mid2026.05 | 44.1 | 62.8 | 70.3 | 55.2 | 71.2 | 77.6 | 20.1 | 42.4 | 47.5 | |
| Single PromptModel=Qwen-Image, Mask=false, Layers / Steps=none2026.05 | 44 | 68.2 | 64.5 | 51.8 | 73.7 | 70.3 | 12.5 | 40.5 | 30.9 | |
| Single PromptModel=FLUX.2-klein, Mask=false, Layers / Steps=none2026.05 | 43.7 | 67.3 | 65 | 53.3 | 74 | 72.1 | 16.9 | 46.1 | 36.7 | |
| masked activation-transport mechanismModel=FLUX.1-dev, Mask=true, Layers / Steps=mid2026.05 | 42.5 | 69.2 | 61.5 | 58.1 | 82.6 | 70.4 | 19.2 | 61.4 | 31.3 | |
| Latent InterpolationModel=FLUX.1-schnell, Mask=false, Layers / Steps=all2026.05 | 42.2 | 63.7 | 66.2 | 52.2 | 71.3 | 73.2 | 18.5 | 39.8 | 46.5 | |
| masked activation-transport mechanismModel=FLUX.2-klein, Mask=true, Layers / Steps=mid2026.05 | 41.7 | 64.6 | 64.5 | 55.2 | 75.2 | 73.4 | 19.1 | 58 | 32.9 | |
| masked activation-transport mechanismModel=SANA1.5, Mask=true, Layers / Steps=mid2026.05 | 41.4 | 72.8 | 56.8 | 55.7 | 83.7 | 66.6 | 17 | 65 | 26.1 | |
| InterpolationModel=FLUX.2-klein, Mask=true, Layers / Steps=mid2026.05 | 41 | 69.8 | 58.8 | 55.2 | 81.9 | 67.4 | 16.4 | 68.4 | 24 | |
| InterpolationModel=FLUX.1-schnell, Mask=true, Layers / Steps=mid2026.05 | 37.4 | 79 | 47.3 | 53.3 | 90.7 | 58.8 | 9.3 | 80.1 | 11.6 | |
| MAsModel=SANA1.5, Mask=false, Layers / Steps=mid2026.05 | 33 | 83 | 39.8 | 56.1 | 98.3 | 57.1 | 5.3 | 96.9 | 5.5 | |
| InterpolationModel=SANA1.5, Mask=true, Layers / Steps=mid2026.05 | 32.7 | 59.2 | 55.3 | 35.5 | 61.4 | 57.9 | 4 | 23.6 | 17 | |
| InterpolationModel=FLUX.1-dev, Mask=true, Layers / Steps=mid2026.05 | 32.6 | 78.4 | 41.6 | 49.6 | 88.4 | 56.1 | 3.2 | 76.5 | 4.2 | |
| Latent InterpolationModel=SANA1.5, Mask=false, Layers / Steps=all2026.05 | 32.5 | 83.4 | 39 | 56.5 | 100 | 56.5 | 5.4 | 99.9 | 5.4 | |
| MAsModel=FLUX.1-schnell, Mask=false, Layers / Steps=mid2026.05 | 32.4 | 83 | 39 | 53 | 98 | 53.4 | 3 | 96.3 | 3.1 | |
| MAsModel=Qwen-Image, Mask=false, Layers / Steps=mid2026.05 | 32.4 | 82.8 | 39.2 | 55.2 | 98.2 | 56.2 | 4 | 96 | 4.2 | |
| InterpolationModel=FLUX.1-schnell, Mask=true, Layers / Steps=all2026.05 | 31.7 | 83.2 | 38.1 | 53.2 | 99.5 | 53.5 | 2.7 | 99 | 2.7 | |
| InterpolationModel=FLUX.1-dev, Mask=true, Layers / Steps=all2026.05 | 31.2 | 74 | 42.2 | 43.4 | 79.7 | 54.5 | 2.4 | 62.4 | 3.8 | |
| Latent InterpolationModel=Qwen-Image, Mask=false, Layers / Steps=all2026.05 | 31.2 | 82.8 | 37.7 | 55.8 | 100 | 55.8 | 3.7 | 99.9 | 3.7 | |
| Latent InterpolationModel=FLUX.2-klein, Mask=false, Layers / Steps=all2026.05 | 31.1 | 81 | 38.4 | 52.4 | 98 | 53.5 | 3.6 | 95.4 | 3.7 | |
| MAsModel=FLUX.1-dev, Mask=false, Layers / Steps=mid2026.05 | 30.9 | 80.4 | 38.5 | 57 | 99.9 | 57 | 4.3 | 99.9 | 4.3 | |
| MAsModel=FLUX.2-klein, Mask=false, Layers / Steps=mid2026.05 | 30.8 | 81 | 38 | 52.7 | 98.7 | 53.4 | 3.3 | 98.6 | 3.4 | |
| Latent InterpolationModel=FLUX.1-dev, Mask=false, Layers / Steps=all2026.05 | 30.6 | 80.2 | 38.1 | 57.1 | 100 | 57.1 | 4.4 | 100 | 4.4 | |
| InterpolationModel=FLUX.2-klein, Mask=true, Layers / Steps=all2026.05 | 30.2 | 81 | 37.3 | 52.7 | 99.1 | 53.2 | 3.2 | 98.7 | 3.2 | |
| InterpolationModel=SANA1.5, Mask=true, Layers / Steps=all2026.05 | 30.2 | 63.8 | 47.3 | 34.1 | 63.7 | 53.6 | 1.9 | 32.6 | 5.8 | |
| InterpolationModel=Qwen-Image, Mask=true, Layers / Steps=all2026.05 | 23.2 | 49.5 | 46.9 | 27.9 | 53.7 | 52.1 | 0 | 2.7 | 1.3 | |
| InterpolationModel=Qwen-Image, Mask=true, Layers / Steps=mid2026.05 | 22.8 | 48.3 | 46.8 | 26.2 | 51.9 | 50.5 | 0 | 2 | 1.3 |