Multi-Reference Image Generation on DreamBench++
68CP (Animal)Qwen-2511 + Ours
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen-2511 + OursBase model=Qwen-2511, Parameter count=20B, Evaluator=GPT-4o2026.06 | 68 | 61 | 69 | 70 | 68 | 90 | 94 | 93 | 92 | 62 | |
| FLUX.2 [klein] + OursBase model=FLUX.2 [klein], Parameter count=9B, Evaluator=GPT-4o2026.06 | 65 | 58 | 66 | 61 | 64 | 94 | 97 | 93 | 95 | 61 | |
| DreamOmni2Evaluator=GPT-4o2026.06 | 64 | 59 | 67 | 53 | 63 | 81 | 80 | 70 | 78 | 49 | |
| FLUX.2 [klein]Parameter count=9B, Evaluator=GPT-4o2026.06 | 59 | 53 | 61 | 54 | 58 | 97 | 97 | 94 | 97 | 56 | |
| Qwen-2511Parameter count=20B, Evaluator=GPT-4o2026.06 | 58 | 55 | 59 | 54 | 58 | 90 | 94 | 93 | 92 | 53 | |
| OmniGen2Evaluator=GPT-4o2026.06 | 54 | 41 | 55 | 49 | 52 | 95 | 93 | 87 | 92 | 48 | |
| BAGELEvaluator=GPT-4o2026.06 | 45 | 23 | 38 | 46 | 39 | 97 | 97 | 91 | 95 | 37 |