Text-to-Image Generation on GenAI-Bench
0.939Basic ScoreSIDiffAgent
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SIDiffAgenttraining_config=ep22026.02 | 0.939 | 0.884 | 0.94 | 0.942 | 0.939 | 0.934 | 0.947 | 0.89 | 0.833 | 0.879 | 0.837 | 0.836 | — | 0.771 | — | — | — | |
| SIDiffAgent2026.02 | 0.918 | 0.852 | 0.919 | 0.924 | 0.918 | 0.909 | 0.928 | 0.857 | 0.805 | 0.855 | 0.797 | 0.796 | — | 0.715 | — | — | — | |
| Recraft v32026.02 | 0.913 | 0.811 | 0.914 | 0.913 | 0.913 | 0.901 | 0.913 | 0.806 | 0.772 | 0.797 | 0.761 | 0.725 | — | 0.589 | — | — | — | |
| Qwen-Agents_v+negvisual_and_negative_feedback=true2026.02 | 0.913 | 0.842 | 0.913 | 0.912 | 0.913 | 0.905 | 0.929 | 0.841 | 0.795 | 0.838 | 0.783 | 0.782 | — | 0.7 | — | — | — | |
| Imagen 3 v0022026.02 | 0.912 | 0.839 | 0.909 | 0.923 | 0.909 | 0.903 | 0.918 | 0.841 | 0.795 | 0.841 | 0.788 | 0.776 | — | 0.673 | 0.914 | — | — | |
| Qwen-Agents2026.02 | 0.907 | 0.789 | 0.906 | 0.912 | 0.91 | 0.897 | 0.921 | 0.79 | 0.777 | 0.814 | 0.742 | 0.69 | — | 0.497 | — | — | — | |
| DALL-E 32026.02 | 0.9 | 0.791 | 0.91 | 0.9 | 0.92 | 0.89 | 0.91 | 0.82 | 0.82 | 0.78 | 0.8 | 0.7 | — | 0.48 | — | — | — | |
| DALL-E 3Params=–, Type=Diff, #Training Images=-, Res.=10242025.03 | 0.9 | — | — | — | — | — | — | — | — | — | — | — | 0.7 | — | — | — | — | |
| T2I-Copilot2026.02 | 0.892 | 0.813 | 0.893 | 0.909 | 0.893 | 0.885 | 0.899 | 0.813 | 0.759 | 0.807 | 0.766 | 0.747 | — | 0.659 | — | — | — | |
| SD 3.5 large2026.02 | 0.89 | 0.764 | 0.891 | 0.895 | 0.889 | 0.88 | 0.895 | 0.76 | 0.743 | 0.763 | 0.707 | 0.659 | — | 0.471 | — | — | — | |
| FLUX1.1-pro2026.02 | 0.884 | 0.766 | 0.89 | 0.899 | 0.884 | 0.871 | 0.894 | 0.766 | 0.751 | 0.788 | 0.71 | 0.666 | — | 0.49 | — | — | — | |
| Qwen-Agent_negnegative_feedback=true2026.02 | 0.884 | 0.818 | 0.883 | 0.9 | 0.885 | 0.882 | 0.888 | 0.824 | 0.777 | 0.82 | 0.76 | 0.764 | — | 0.684 | — | — | — | |
| Lumina-Image-2.02026.02 | 0.874 | 0.752 | 0.879 | 0.896 | 0.876 | 0.872 | 0.885 | 0.76 | 0.729 | 0.767 | 0.723 | 0.649 | — | 0.451 | — | — | — | |
| Midjourney v62026.02 | 0.87 | 0.772 | 0.88 | 0.87 | 0.87 | 0.87 | 0.91 | 0.78 | 0.79 | 0.78 | 0.76 | 0.69 | — | 0.5 | — | — | — | |
| FLUX.1-dev2026.02 | 0.864 | 0.745 | 0.873 | 0.875 | 0.862 | 0.853 | 0.875 | 0.747 | 0.733 | 0.756 | 0.711 | 0.646 | — | 0.456 | — | — | — | |
| Janus Pro-7B2026.02 | 0.859 | 0.747 | 0.865 | 0.886 | 0.867 | 0.856 | 0.87 | 0.731 | 0.734 | 0.759 | 0.693 | 0.653 | — | 0.48 | — | — | — | |
| Qwen-Image2026.02 | 0.853 | 0.757 | 0.857 | 0.874 | 0.858 | 0.853 | 0.861 | 0.766 | 0.753 | 0.777 | 0.738 | 0.677 | — | 0.501 | — | — | — | |
| UniTokParams=7B, Type=AR, #Training Images=30M, Res.=2562025.03 | 0.85 | — | — | — | — | — | — | — | — | — | — | — | 0.67 | — | — | — | — | |
| SDXLParams.=2.6B, Type=Diffusion2024.12 | 0.83 | — | — | — | — | — | — | — | — | — | — | — | 0.63 | — | — | — | — | |
| SDXLParams=2.6B, Type=Diff, #Training Images=2000M, Res.=10242025.03 | 0.83 | — | — | — | — | — | — | — | — | — | — | — | 0.63 | — | — | — | — | |
| LiquidParams=7B, Type=AR, #Training Images=30M, Res.=5122025.03 | 0.83 | — | — | — | — | — | — | — | — | — | — | — | 0.65 | — | — | — | — | |
| SemHiTokParams=7B, Type=AR, #Training Images=15M, Res.=2562025.03 | 0.83 | — | — | — | — | — | — | — | — | — | — | — | 0.64 | — | — | — | — | |
| SemHiTokParams=7B, Type=AR, #Training Images=15M, Res.=3842025.03 | 0.83 | — | — | — | — | — | — | — | — | — | — | — | 0.66 | — | — | — | — | |
| UniTokLLM=LLaMA-2-7B, PT Data Size=70M2025.05 | 0.82 | — | — | — | — | — | — | — | — | — | — | — | 0.66 | — | — | — | — | |
| Kolors v1.02026.02 | 0.819 | 0.711 | 0.821 | 0.841 | 0.832 | 0.818 | 0.803 | 0.737 | 0.705 | 0.726 | 0.695 | 0.621 | — | 0.438 | — | — | — | |
| HunyuanDiT v1.22026.02 | 0.818 | 0.721 | 0.817 | 0.855 | 0.825 | 0.827 | 0.798 | 0.732 | 0.743 | 0.723 | 0.692 | 0.64 | — | 0.475 | — | — | — | |
| Playground v2.52026.02 | 0.815 | 0.72 | 0.818 | 0.85 | 0.803 | 0.818 | 0.821 | 0.732 | 0.721 | 0.696 | 0.695 | 0.64 | — | 0.499 | — | — | — | |
| SD v2.1Params=–, Type=Diff, #Training Images=2000M, Res.=10242025.03 | 0.78 | — | — | — | — | — | — | — | — | — | — | — | 0.62 | — | — | — | — | |
| LaVITLLM=LLaMA-7B, PT Data Size=100M2025.05 | 0.78 | — | — | — | — | — | — | — | — | — | — | — | 0.62 | — | — | — | — | |
| VILA-U (256)Params.=7B, Type=Autoregressive2024.12 | 0.76 | — | — | — | — | — | — | — | — | — | — | — | 0.64 | — | — | — | — | |
| VILA-UParams=7B, Type=AR, #Training Images=15M, Res.=2562025.03 | 0.76 | — | — | — | — | — | — | — | — | — | — | — | 0.64 | — | — | — | — | |
| VILA-ULLM=LLaMA-2-7B, PT Data Size=23M2025.05 | 0.76 | — | — | — | — | — | — | — | — | — | — | — | 0.64 | — | — | — | — | |
| ILLUMEParams.=7B, Type=Autoregressive2024.12 | 0.75 | — | — | — | — | — | — | — | — | — | — | — | 0.6 | — | — | — | — | |
| ILLUMEParams=7B, Type=AR, #Training Images=15M, Res.=5122025.03 | 0.75 | — | — | — | — | — | — | — | — | — | — | — | 0.6 | — | — | — | — | |
| ILLUME+Params=7B, Type=AR, #Training Images=15M, Res.=5122025.03 | 0.75 | — | — | — | — | — | — | — | — | — | — | — | 0.6 | — | — | — | — | |
| VILA-U (384)Params.=7B, Type=Autoregressive2024.12 | 0.73 | — | — | — | — | — | — | — | — | — | — | — | 0.61 | — | — | — | — | |
| VILA-UParams=7B, Type=AR, #Training Images=15M, Res.=3842025.03 | 0.73 | — | — | — | — | — | — | — | — | — | — | — | 0.61 | — | — | — | — | |
| Show-oParams.=1.5B, Type=Autoregressive2024.12 | 0.7 | — | — | — | — | — | — | — | — | — | — | — | 0.6 | — | — | — | — | |
| Show-oParams=1.5B, Type=Diff, #Training Images=36M, Res.=2562025.03 | 0.7 | — | — | — | — | — | — | — | — | — | — | — | 0.6 | — | — | — | — | |
| GenArtist2026.02 | 0.693 | 0.588 | 0.702 | 0.736 | 0.659 | 0.677 | 0.688 | 0.553 | 0.518 | 0.473 | 0.546 | 0.504 | — | 0.437 | — | — | — | |
| Slot-MLLM (7B)LLM=Vicuna-7B, PT Data Size=25M2025.05 | 0.67 | — | — | — | — | — | — | — | — | — | — | — | 0.59 | — | — | — | — | |
| TokenFlowParams=7B, Type=AR, #Training Images=60M, Res.=2562025.03 | 0.65 | — | — | — | — | — | — | — | — | — | — | — | 0.65 | — | — | — | — | |
| ChameleonParams.=7B, Type=Autoregressive2024.12 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | 0.53 | — | — | — | — | |
| LWMParams.=7B, Type=Autoregressive2024.12 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | 0.53 | — | — | — | — | |
| LWMParams=7B, Type=AR, #Training Images=-, Res.=2562025.03 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | 0.53 | — | — | — | — | |
| Slot-MLLM (14B)LLM=Qwen2.5-14B-Instruct, PT Data Size=25M2025.05 | 0.58 | — | — | — | — | — | — | — | — | — | — | — | 0.54 | — | — | — | — | |
| SEED-LLaMALLM=Vicuna-7B, PT Data Size=77M2025.05 | 0.52 | — | — | — | — | — | — | — | — | — | — | — | 0.49 | — | — | — | — | |
| B1: Penult.Fusion Strategy=Penultimate-layer, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 74.96 | — | |
| B2: UniformFusion Strategy=Uniform normalized averaging, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 76.82 | — | |
| B3: StaticFusion Strategy=Learnable static fusion, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 76.31 | — | |
| bge-Gemma2Parameters=9B, Embeddings=last layer2025.06 | — | 0.737 | 0.73 | 0.824 | 0.729 | 0.793 | 0.722 | 0.662 | 0.654 | 0.641 | 0.623 | 0.797 | — | — | — | — | — | |
| bge-Gemma2Parameters=9B, Embeddings=avg2025.06 | — | 0.774 | 0.776 | 0.851 | 0.765 | 0.813 | 0.752 | 0.716 | 0.773 | 0.702 | 0.665 | 0.822 | — | — | — | — | — | |
| bge-Gemma2Parameters=9B, Embeddings=norm avg2025.06 | — | 0.789 | 0.787 | 0.846 | 0.782 | 0.821 | 0.786 | 0.745 | 0.776 | 0.744 | 0.712 | 0.81 | — | — | — | — | — | |
| DDPOFine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 65.7 | |
| DDPOFine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 64.65 | |
| Flux-VAEModel=DiTDH-XL, Params=0.9B, Protocol=Pretraining2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 57.3 | — | |
| Flux-VAEModel=DiTDH-XL, Params=0.9B, Protocol=Finetuning2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 63.9 | — | |
| FuseDiTFusion Strategy=Deep-fusion baseline, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 75.02 | — | |
| GRAFTFine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 70.51 | |
| GRAFTFine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 70.97 | |
| GRAN-TED2025.12 | — | 77.41 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mistral-7BParameters=7B, Embeddings=last layer2025.06 | — | 0.675 | 0.667 | 0.763 | 0.665 | 0.711 | 0.641 | 0.576 | 0.556 | 0.526 | 0.524 | 0.726 | — | — | — | — | — | |
| Mistral-7BParameters=7B, Embeddings=avg2025.06 | — | 0.731 | 0.733 | 0.806 | 0.731 | 0.779 | 0.712 | 0.683 | 0.666 | 0.639 | 0.582 | 0.758 | — | — | — | — | — | |
| Mistral-7BParameters=7B, Embeddings=norm avg2025.06 | — | 0.769 | 0.774 | 0.837 | 0.78 | 0.802 | 0.733 | 0.699 | 0.716 | 0.706 | 0.63 | 0.789 | — | — | — | — | — | |
| Norm-Avg2025.12 | — | 76.17 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| P-GRAFT(0.25N)Fine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.94 | |
| P-GRAFT(0.25N)Fine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.42 | |
| P-GRAFT(0.5N)Fine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71 | |
| P-GRAFT(0.5N)Fine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 70.73 | |
| P-GRAFT(0.75N)Fine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 69.46 | |
| P-GRAFT(0.75N)Fine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 69.51 | |
| RAEModel=DiTDH-XL, Params=0.9B, Protocol=Pretraining2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 63.2 | — | |
| RAEModel=DiTDH-XL, Params=0.9B, Protocol=Finetuning2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 67.2 | — | |
| RAEv2Model=DiTDH-XL, Params=0.9B, Protocol=Pretraining2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 63.8 | — | |
| RAEv2Model=DiTDH-XL, Params=0.9B, Protocol=Finetuning2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 68 | — | |
| S1: TimeFusion Strategy=Time-wise fusion, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 76.2 | — | |
| S2: DepthFusion Strategy=Depth-wise fusion, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 79.07 | — | |
| S3: JointFusion Strategy=Joint fusion, Judge Model=Qwen3-VL-235B-A22B-Instruct, Sampling steps=50, CFG scale=6.0, Scheduler=FlowMatch Euler2026.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 77.44 | — | |
| SD v2Fine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 66.87 | |
| SD v2Fine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 66.87 | |
| SDXL-BaseFine-tuning dataset=GenAI-Bench2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 69.69 | |
| SDXL-BaseFine-tuning dataset=T2ICompBench++ - Train2025.10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 69.69 | |
| T5-XXLParameters=4.7B, Embeddings=last layer2025.06 | — | 0.741 | 0.737 | 0.809 | 0.741 | 0.782 | 0.723 | 0.677 | 0.717 | 0.675 | 0.599 | 0.757 | — | — | — | — | — | |
| T5-XXLParameters=4.7B, Embeddings=norm avg2025.06 | — | 0.747 | 0.748 | 0.813 | 0.745 | 0.78 | 0.72 | 0.687 | 0.736 | 0.675 | 0.617 | 0.76 | — | — | — | — | — |