Subject-driven Image Generation on DreamBench
81.6DINO ScoreUNIC-Adapter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| UNIC-AdapterEvaluation Protocol=Test Time Tuning Free2024.12 | 81.6 | 84.1 | 30.6 | — | |
| OmniGenEvaluation Protocol=Test Time Tuning Free2024.12 | 80.1 | 84.7 | 30.1 | — | |
| Real Image (Oracle)2023.04 | 77.4 | 88.5 | — | — | |
| Real Images (Oracle)Evaluation Protocol=Oracle2023.11 | 77.4 | 88.5 | — | — | |
| Reference ImagesBackbone=N/A, Iterations=N/A2024.07 | 77.4 | 88.5 | — | — | |
| Oracleinput=reference images2026.04 | 77.4 | 88.5 | — | — | |
| MUSIC*test-time scaling=true2026.04 | 76.8 | 84 | 32.1 | — | |
| DreamVARBackbone=Infinity-2B, Parameters=2B, Resolution=512 x 5122026.01 | 76.4 | 83.8 | 31 | — | |
| MUSIC2026.04 | 76.1 | 83.7 | 31.7 | — | |
| UNO2026.04 | 76 | 83.5 | 30.4 | — | |
| EchoGen-2BBase Model=Infinity-2B, Evaluation protocol=Feed-Forward, Latency=5.2s2025.09 | 75.5 | 83.5 | 32.5 | — | |
| DreamOArchitecture type=VAE-based2026.05 | 75.37 | 83.56 | 30.86 | — | |
| AR-BoothBase Model=Infinity-2B, Evaluation protocol=Test-time Fine-tuning, Latency=2.8h2025.09 | 75 | 80.8 | 26.9 | — | |
| UNOArchitecture type=VAE-based2026.05 | 74.84 | 83.54 | 30.4 | — | |
| Ours (MLLM + VAE)Architecture type=MLLM-based, Conditioning=Full model (MLLM + VAE), Training data=Public data (UNO-1M)2026.05 | 74.82 | 84.43 | 30.1 | — | |
| UMOArchitecture type=VAE-based2026.05 | 74.81 | 83.39 | 30.22 | — | |
| USOArchitecture type=VAE-based2026.05 | 74.78 | 82.63 | 32.13 | — | |
| UNOParameters=12B2026.01 | 74.7 | 82.9 | 30.3 | — | |
| SuTIBackbone=Imagen [1]2023.04 | 74.1 | 81.9 | 0.304 | — | |
| SuTIEvaluation Protocol=Test Time Tuning Free2024.12 | 74.1 | 81.9 | 30.4 | — | |
| SuTIBackbone=Imagen, Iterations=1.5 × 10^52024.07 | 74.1 | 81.9 | 30.4 | — | |
| OmniGen2Architecture type=VAE-based2026.05 | 73.23 | 82.68 | 31.85 | — | |
| Qwen-ImageArchitecture type=MLLM-based2026.05 | 73.17 | 82.61 | 31.58 | — | |
| XVerseArchitecture type=VAE-based2026.05 | 72.15 | 81.75 | 30.98 | — | |
| UNO*Training Data=Same as DreamVAR, Parameters=12B2026.01 | 71.7 | 81.9 | 30.2 | — | |
| Subject-Diff2024.11 | 71.1 | 78.7 | 30.3 | — | |
| CoDi-2Evaluation Protocol=Test Time Tuning Free2023.11 | 70.3 | 85.2 | 31.1 | — | |
| RealCustom++2026.04 | 70.2 | 79.4 | 31.8 | — | |
| EasyRefArchitecture type=MLLM-based2026.05 | 69.61 | 81.53 | 30.31 | — | |
| DreamBoothBackbone=Imagen [1]2023.04 | 69.6 | 81.2 | 0.306 | — | |
| DreamBoothBackbone=Imagen, Iterations=10002024.07 | 69.6 | 81.2 | 30.6 | — | |
| IP-Adapter2026.01 | 69.6 | 80.7 | 27.8 | — | |
| Custom DiffusionBackbone=SD, Iterations=5002024.07 | 69.5 | 80.1 | 24.5 | — | |
| KOSMOS-GEvaluation Protocol=Test Time Tuning Free2023.11 | 69.4 | 84.7 | 28.7 | — | |
| Kosmos-GEvaluation Protocol=Test Time Tuning Free2024.12 | 69.4 | 84.7 | 28.7 | — | |
| Kosmos-GModel=SD-v1.52024.11 | 69.4 | 84.7 | 28.7 | — | |
| OmniGen2026.01 | 69.3 | 80.1 | 31.5 | — | |
| OmniGenBase Model=OmniGen, Evaluation protocol=Unified Generation, Latency=93.4s2025.09 | 69.3 | 80.1 | 31.5 | — | |
| OmniGen2026.04 | 69.3 | 80.1 | 31.5 | — | |
| Diptych PromptingModel=FLUX2024.11 | 68.9 | 75.8 | 34.4 | — | |
| UNOArchitecture type=VAE-based, Training data=UNO-1M (public), Special notation=†2026.05 | 68.6 | 81.61 | 30.71 | — | |
| OminiControlParameters=12B2026.01 | 68.4 | 79.9 | 31.2 | — | |
| OminiControlBase Model=FLUX.1-dev, Evaluation protocol=Feed-Forward, Latency=27.5s2025.09 | 68.4 | 79.9 | 31.2 | — | |
| OminiControl2026.04 | 68.4 | 79.9 | 31.2 | — | |
| Ours (MLLM only)Architecture type=MLLM-based, Stage=First stage only, Framework=MLLM-DiT2026.05 | 67.88 | 82.28 | 29.88 | — | |
| EchoGen-0.1BBase Model=Infinity-0.1B, Evaluation protocol=Feed-Forward, Latency=0.5s2025.09 | 67.5 | 80.6 | 32.1 | — | |
| BootPIG2026.01 | 67.4 | 79.7 | 31.1 | — | |
| BootPIG2026.04 | 67.4 | 79.7 | 31.1 | — | |
| ELITE-SAGSAG=true2024.05 | 67.1 | 79 | 34.4 | — | |
| MS-DiffModel=SD-XL2024.11 | 67.1 | 79.2 | 32.1 | — | |
| MS-DiffusionBase Model=SDXL, Evaluation protocol=Feed-Forward, Latency=39.6s2025.09 | 67.1 | 79.2 | 32.1 | — | |
| MS-Diffusion2024.12 | 67.1 | 79.2 | 32.1 | — | |
| BLIP-DiffusionEvaluation Protocol=Fine-Tuning2024.12 | 67 | 80.5 | 30.2 | — | |
| BLIP-DiffusionBase Model=SD-v1.5, Evaluation protocol=Test-time Fine-tuning2025.09 | 67 | 80.5 | 30.2 | — | |
| BLIP-Diffusion2026.04 | 67 | 80.5 | 30.2 | — | |
| DreamBoothBackbone=SD [21]2023.04 | 66.8 | 80.3 | 0.305 | — | |
| DreamBoothEvaluation Protocol=Fine-Tuning2023.11 | 66.8 | 80.3 | 30.5 | — | |
| DreamboothType=Finetune-based2023.12 | 66.8 | 80.3 | 30.5 | 6.33 | |
| DreamBoothEvaluation Protocol=Fine-Tuning2024.12 | 66.8 | 80.3 | 30.5 | — | |
| DreamBoothBackbone=SD, Iterations=10002024.07 | 66.8 | 80.3 | 30.5 | — | |
| DreamBooth2026.01 | 66.8 | 80.3 | 30.5 | — | |
| DreamBoothBase Model=SD-v1.5, Evaluation protocol=Test-time Fine-tuning, Latency=15min2025.09 | 66.8 | 80.3 | 30.5 | — | |
| DreamBooth2026.04 | 66.8 | 80.3 | 30.5 | — | |
| Break-A-SceneType=Finetune-based2023.12 | 65.3 | 78.8 | 28.7 | 6.234 | |
| ELETEBackbone=SD, Iterations=30002024.07 | 65.2 | 76.5 | 25.5 | — | |
| RPOBackbone=SD, Iterations=4002024.07 | 65.2 | 83.3 | 31.4 | — | |
| EasyControlBase Model=FLUX.1-dev, Evaluation protocol=Feed-Forward, Latency=25.4s2025.09 | 65.2 | 78.9 | 32.5 | — | |
| DreamBooth2024.05 | 65.1 | 78.5 | 31.5 | — | |
| ELITE2026.01 | 64.7 | 77.2 | 29.6 | — | |
| ELITE2026.04 | 64.7 | 77.2 | 29.6 | — | |
| ELITEModel=SD-v1.42024.11 | 62.1 | 77.1 | 29.3 | — | |
| ELITEBase Model=SD-v1.4, Evaluation protocol=Feed-Forward, Latency=11.0s2025.09 | 62.1 | 77.1 | 29.3 | — | |
| OmniPrism2024.12 | 61.4 | 81.5 | 32.8 | — | |
| λ-EclipseModel=Kan-v2.22024.11 | 61.3 | 78.3 | 30.7 | — | |
| IP-AdapterModel=SD-XL, Evaluation source=values from [33]2024.11 | 61.3 | 81 | 29.2 | — | |
| λ-EclipseBase Model=Kan-v2.2, Evaluation protocol=Feed-Forward2025.09 | 61.3 | 78.3 | 30.7 | — | |
| IP-AdapterBase Model=SDXL, Evaluation protocol=Feed-Forward, Latency=16.9s2025.09 | 61.3 | 81 | 29.2 | — | |
| SSR-EncoderType=Finetune-free2023.12 | 61.2 | 82.1 | 30.8 | 6.563 | |
| SSR-EncoderBackbone=SD, Iterations=10^62024.07 | 61.2 | 82.1 | 31.4 | — | |
| SSR-Encoder2026.04 | 61.2 | 82.1 | 30.8 | — | |
| SSR-Encoder2024.12 | 61.2 | 82.1 | 30.8 | — | |
| IP-adapterType=Finetune-free2023.12 | 60.8 | 80.9 | 27.4 | 6.432 | |
| IP-AdapterBackbone=SD, Iterations=10^62024.07 | 60.8 | 80.9 | 27.4 | — | |
| IP-Adapter2024.12 | 60.8 | 80.9 | 27.4 | — | |
| ELITEType=Finetune-free2023.12 | 60.5 | 77.5 | 29.8 | 6.283 | |
| Re-ImagenBackbone=Imagen [1]2023.04 | 60 | 74 | 0.27 | — | |
| Re-ImagenEvaluation Protocol=Test Time Tuning Free2023.11 | 60 | 74 | 27 | — | |
| Re-ImagenEvaluation Protocol=Test Time Tuning Free2024.12 | 60 | 74 | 27 | — | |
| Re-ImagenBackbone=Imagen, Iterations=2 × 10^52024.07 | 60 | 74 | 27 | — | |
| Re-ImagenBase Model=Imagen, Evaluation protocol=Feed-Forward2025.09 | 60 | 74 | 27 | — | |
| Re-Imagen2026.04 | 60 | 74 | 27 | — | |
| OminiControlArchitecture type=VAE-based2026.05 | 59.87 | 78.4 | 31.86 | — | |
| BLIP-DiffusionType=Finetune-free2023.12 | 59.4 | 77.9 | 30 | 6.212 | |
| BLIP-DiffModel=SD-v1.52024.11 | 59.4 | 77.9 | 30 | — | |
| BLIP-DiffusionBase Model=SD-v1.5, Evaluation protocol=Feed-Forward2025.09 | 59.4 | 77.9 | 30 | — | |
| BLIP-Diffusion2024.12 | 59.4 | 77.9 | 30 | — | |
| ELITESAG=false2024.05 | 58.6 | 75.1 | 34.2 | — | |
| Flux.1 IP-Adapter2026.01 | 58.2 | 82 | 28.8 | — | |
| FLUX.1 IP-Adapter2026.04 | 58.2 | 82 | 28.8 | — | |
| DisenBoothBackbone=SD, Iterations=30002024.07 | 57.4 | 75.5 | 25.5 | — |