Multi-reference image generation on MultiBanana
6.52Fidelity (Single Object)DreamOmni2
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DreamOmni2Evaluator=Gemini-2.5 and GPT-5, Scale=10-point2026.06 | 6.52 | 4.07 | 2.8 | 3.04 | 2.87 | 2.59 | 3.65 | |
| Qwen-2511Evaluator=Gemini-2.5 and GPT-5, Scale=10-point2026.06 | 6.37 | 4.54 | 3.01 | 2.93 | 3.18 | 2.65 | 3.78 | |
| Qwen-2511 + DyRefEvaluator=Gemini-2.5 and GPT-5, Scale=10-point2026.06 | 6.34 | 4.42 | 3.96 | 3.65 | 4.04 | 3.63 | 4.34 | |
| OmniGen2Evaluator=Gemini-2.5 and GPT-5, Scale=10-point2026.06 | 5.92 | 3.44 | 3.26 | 3.6 | 3.37 | 3.02 | 3.77 |