Text-to-image concept steering on 167 held-out concepts (test)
83.1Stage 1 Input Fidelity (VLM)Prompting
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PromptingModel=DMD22026.05 | 83.1 | 57.7 | 30.5 | 29.7 | |
| PromptingModel=Nitro-1-PixArt2026.05 | 82.7 | 51.7 | 37 | 29.6 | |
| HyperTransportModel=Nitro-1-PixArt2026.05 | 71.1 | 39.9 | 63 | 70.4 | |
| HyperTransportModel=DMD22026.05 | 70.1 | 43.7 | 69.5 | 70.3 |