Text-to-Image Generation on GenEval 1.0 (test)
95Overall ScoreSD3.5-M w/ Flow-GRPO
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SD3.5-M w/ Flow-GRPOModel Type=Diffusion, Optimization=Flow-GRPO2026.05 | 95 | 100 | 99 | 95 | 92 | 99 | — | 86 | — | — | — | — | |
| SD3.5-M w/ RTDMDModel Type=Few-Step Diffusion, NFE=42026.05 | 94 | 100 | 98 | 95 | 90 | 87 | — | 94 | — | — | — | — | |
| SD3.5-M w/ TDM-R1Model Type=Few-Step Diffusion, NFE=42026.05 | 92 | 100 | 96 | 88 | 85 | 93 | — | 91 | — | — | — | — | |
| DeCo-XXL/16#Params=1.1B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 86 | 100 | 92 | 72 | 91 | 80 | 79 | — | — | — | — | — | |
| RaPD#Params=1.1B, Resolution=512 x 5122026.05 | 85 | 99 | 93 | 72 | 93 | 77 | 74 | — | — | — | — | — | |
| BLIP3-o 8BModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 84 | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4oModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 84 | 99 | 92 | 85 | 92 | 75 | — | 61 | — | — | — | — | |
| GvUModel Type=Generation & Understanding, LLM prompt rewriter=true2026.03 | 84 | 100 | 97 | 80 | 93 | 68 | — | 65 | — | — | — | — | |
| GPT-4oModel Type=Autoregressive2026.05 | 84 | 99 | 92 | 85 | 92 | 75 | — | 61 | — | — | — | — | |
| DeCo-XXL/16re-run=true, #Params=1.1B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 83 | 99 | 92 | 65 | 91 | 71 | 77 | — | — | — | — | — | |
| Self-Evaluating ModelSteps=50 Steps2025.12 | 81.51 | 98.75 | 93.94 | 70 | — | 70.25 | — | 67 | 89.1 | — | — | — | |
| BAGELModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 81 | 99 | 94 | 80 | 87 | 64 | — | 63 | — | — | — | — | |
| GvUModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 81 | 100 | 96 | 74 | 92 | 61 | — | 58 | — | — | — | — | |
| SANA-1.5 4.8BModel Type=Diffusion2026.05 | 81 | 99 | 93 | 86 | 84 | 59 | — | 65 | — | — | — | — | |
| SANA-1.5Steps=50 Steps2025.12 | 80.62 | 98.44 | 91.92 | 70.31 | — | 59 | — | 71.75 | 92.29 | — | — | — | |
| Janus-ProModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 80 | 99 | 89 | 59 | 90 | 79 | — | 66 | — | — | — | — | |
| UniWorld-V1Model Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 80 | 99 | 93 | 79 | 89 | 49 | — | 70 | — | — | — | — | |
| OmniGen2Model Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 80 | 100 | 95 | 64 | 88 | 55 | — | 76 | — | — | — | — | |
| Janus-Pro-7BModel Type=Autoregressive2026.05 | 80 | 99 | 89 | 59 | 90 | 79 | — | 66 | — | — | — | — | |
| FLUX.1-DevSteps=50 Steps2025.12 | 79.66 | 97.81 | 93.18 | 75 | — | 65 | — | 56 | 90.96 | — | — | — | |
| PixelGen-XXL/16#Params=1.1B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 79 | 99 | 88 | 59 | 90 | 70 | 70 | — | — | — | — | — | |
| PixelGen-XXL/16re-run=true, #Params=1.1B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 79 | 99 | 88 | 58 | 91 | 70 | 67 | — | — | — | — | — | |
| Self-Evaluating ModelSteps=8 Steps2025.12 | 78.49 | 96.88 | 91.41 | 66.88 | — | 65.25 | — | 62.25 | 88.3 | — | — | — | |
| Self-Evaluating ModelSteps=4 Steps2025.12 | 78.06 | 96.88 | 91.41 | 62.19 | — | 66 | — | 62.5 | 89.36 | — | — | — | |
| TiMSteps=50 Steps2025.12 | 77.97 | 96.56 | 88.64 | 63.44 | — | 55.5 | — | 73 | 90.69 | — | — | — | |
| SANA-1.5Steps=8 Steps2025.12 | 77.88 | 98.12 | 88.64 | 67.5 | — | 63 | — | 58 | 92.02 | — | — | — | |
| Self-Evaluating ModelSteps=2 Steps2025.12 | 75.31 | 98.12 | 88.38 | 60.94 | — | 63.25 | — | 59 | 82.18 | — | — | — | |
| Janus-Pro 1B + VA-πExt. Rwd=false, Backbone=Janus-Pro 1B2025.12 | 74.4 | 98.8 | 83.5 | 54 | 91.2 | 60 | — | 58.5 | — | — | — | — | |
| SD3-mediumModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 74 | 99 | 94 | 72 | 89 | 33 | — | 60 | — | — | — | — | |
| XOmniModel Type=Generation & Understanding, LLM prompt rewriter=true2026.03 | 74 | 99 | 95 | 69 | 90 | 51 | — | 37 | — | — | — | — | |
| SD3-Medium2026.04 | 74 | 99 | 94 | 72 | — | 33 | 60 | — | 89 | — | — | — | |
| Janus-Pro (1B)Base Model=Janus-Pro (1B), Decoding=Autoregressive2025.12 | 73 | 98.62 | 83.24 | 43.92 | 88.67 | 60.73 | 49.81 | — | — | 14.42 | 1 | 1 | |
| PixNerd-XXL/16#Params=1.2B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 73 | 97 | 86 | 44 | 83 | 71 | 53 | — | — | — | — | — | |
| Janus-Pro 1BExt. Rwd=None, Backbone=Janus-Pro 1B2025.12 | 72.5 | 97.2 | 80.1 | 53.1 | 90.2 | 60.5 | — | 54 | — | — | — | — | |
| SD3.5-TurboSteps=4 Steps2025.12 | 71.94 | 93.44 | 85.1 | 56.56 | — | 60.5 | — | 56.5 | 79.52 | — | — | — | |
| TiMSteps=8 Steps2025.12 | 71.43 | 96.56 | 82.32 | 51.56 | — | 51.25 | — | 57.5 | 89.36 | — | — | — | |
| Janus-Pro (1B) + SJDBase Model=Janus-Pro (1B), Decoding=Speculative Jacobi Decoding2025.12 | 71 | 98.78 | 82.97 | 43.61 | 88.19 | 61.12 | 49.95 | — | — | 7.99 | 1.8 | 1.95 | |
| SD3.5-largeModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 71 | 98 | 89 | 73 | 83 | 34 | — | 47 | — | — | — | — | |
| PixNerd-XXL/16re-run=true, #Params=1.2B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 71 | 98 | 82 | 43 | 81 | 66 | 55 | — | — | — | — | — | |
| SD3.5-LModel Type=Diffusion2026.05 | 71 | 98 | 89 | 73 | 83 | 34 | — | 47 | — | — | — | — | |
| SD3.5-TurboSteps=8 Steps2025.12 | 70.71 | 94.37 | 82.32 | 53.12 | — | 57.25 | — | 54.5 | 82.71 | — | — | — | |
| Janus-Pro (1B) + SJD++Base Model=Janus-Pro (1B), Decoding=SJD with token reuse2025.12 | 70.7 | 98.75 | 82.83 | 43.75 | 88.3 | 61 | 50 | — | — | 6.31 | 2.29 | 2.46 | |
| SD3.5-L-TurboModel Type=Few-Step Diffusion, NFE=42026.05 | 70 | 94 | 84 | 55 | 79 | 58 | — | 56 | — | — | — | — | |
| TiMSteps=4 Steps2025.12 | 68.67 | 95.31 | 76.01 | 50.31 | — | 48 | — | 52.25 | 90.16 | — | — | — | |
| XOmniModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 68 | 100 | 94 | 60 | 85 | 40 | — | 26 | — | — | — | — | |
| SD3#Params=8B, Model Type=Latent Diffusion Model, Resolution=512 x 5122026.05 | 68 | 98 | 84 | 66 | 74 | 40 | 43 | — | — | — | — | — | |
| DALL-EModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 67 | 96 | 87 | 47 | 83 | 43 | — | 45 | — | — | — | — | |
| FLUX.1-devModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 67 | 99 | 81 | 79 | 74 | 20 | — | 47 | — | — | — | — | |
| DALL-E32026.04 | 67 | 96 | 87 | 47 | — | 43 | 45 | — | 83 | — | — | — | |
| FLUX.1-dev#Params=12B, Model Type=Latent Diffusion Model, Resolution=512 x 5122026.05 | 67 | 99 | 81 | 79 | 74 | 20 | 47 | — | — | — | — | — | |
| DALL-E 3Model Type=Latent Diffusion Model, Resolution=512 x 5122026.05 | 67 | 96 | 87 | 47 | 83 | 43 | 45 | — | — | — | — | — | |
| DALLE-3Model Type=Diffusion2026.05 | 67 | 96 | 87 | 47 | 83 | 43 | — | 45 | — | — | — | — | |
| FLUX.1 DevModel Type=Diffusion2026.05 | 66 | 98 | 81 | 74 | 79 | 22 | — | 45 | — | — | — | — | |
| TiMSteps=2 Steps2025.12 | 63.38 | 94.69 | 70.71 | 41.88 | — | 42 | — | 43.75 | 87.23 | — | — | — | |
| PFT-1.2B + look-aheadSampling strategy=look-ahead2026.04 | 63 | 97 | 73 | 41 | — | 35 | 46 | — | 86 | — | — | — | |
| JanusFlowModel Type=Autoregressive2026.05 | 63 | 97 | 59 | 45 | 83 | 53 | — | 42 | — | — | — | — | |
| SD3.5-MModel Type=Diffusion2026.05 | 63 | 98 | 78 | 50 | 81 | 24 | — | 52 | — | — | — | — | |
| SD3.5-TurboSteps=50 Steps2025.12 | 61.14 | 86.56 | 74.49 | 42.81 | — | 52.5 | — | 40.5 | 69.95 | — | — | — | |
| IF-XLModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 61 | 97 | 74 | 66 | 81 | 13 | — | 35 | — | — | — | — | |
| JanusModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 61 | 97 | 68 | 30 | 84 | 46 | — | 42 | — | — | — | — | |
| SD3.5-M w/ TDMModel Type=Few-Step Diffusion, NFE=42026.05 | 61 | 99 | 77 | 49 | 79 | 23 | — | 44 | — | — | — | — | |
| PixelFlow-XL/4#Params=0.9B, Model Type=Pixel Diffusion Model, Resolution=512 x 5122026.05 | 60 | — | — | — | — | — | — | — | — | — | — | — | |
| SimpleAR (0.5B) + SJD++Base Model=SimpleAR (0.5B), Decoding=SJD with token reuse2025.12 | 59.94 | 98.75 | 87.88 | 30 | 82.98 | 25 | 35 | — | — | 29.75 | 2.81 | 3.31 | |
| SimpleAR (0.5B)Base Model=SimpleAR (0.5B), Decoding=Autoregressive2025.12 | 59.06 | 98.32 | 87.12 | 29.84 | 82.61 | 25.47 | 34.92 | — | — | 83.55 | 1 | 1 | |
| PFT-1.2BSampling strategy=standard2026.04 | 59 | 94 | 66 | 44 | — | 32 | 45 | — | 76 | — | — | — | |
| PFT-1.2B + dual-loopSampling strategy=dual-loop2026.04 | 59 | 94 | 69 | 43 | — | 36 | 40 | — | 70 | — | — | — | |
| FLUX.1-DevSteps=8 Steps2025.12 | 58.93 | 88.44 | 72.98 | 46.25 | — | 51 | — | 21.75 | 73.14 | — | — | — | |
| SimpleAR (0.5B) + SJDBase Model=SimpleAR (0.5B), Decoding=Speculative Jacobi Decoding2025.12 | 58.9 | 98.56 | 87.41 | 30.11 | 82.43 | 24.93 | 34.75 | — | — | 42.73 | 1.96 | 2.34 | |
| LlamaGen (0.8B) + SJD++Base Model=LlamaGen (0.8B), Decoding=SJD with token reuse2025.12 | 58.87 | 98 | 73.01 | 31.5 | 81.85 | 21.58 | 28.5 | — | — | 19.45 | 2.3 | 3.25 | |
| LlamaGen (0.8B)Base Model=LlamaGen (0.8B), Decoding=Autoregressive2025.12 | 58.2 | 97.82 | 72.56 | 31.25 | 81.4 | 21.63 | 28.43 | — | — | 44.73 | 1 | 1 | |
| LlamaGen (0.8B) + SJDBase Model=LlamaGen (0.8B), Decoding=Speculative Jacobi Decoding2025.12 | 58.05 | 97.9 | 72.74 | 31.37 | 81.62 | 21.4 | 28.22 | — | — | 31.92 | 1.4 | 2.12 | |
| SANA-1.5Steps=4 Steps2025.12 | 57.25 | 92.19 | 63.13 | 51.25 | — | 42 | — | 25.25 | 69.68 | — | — | — | |
| SD XLModel Type=Only Generation, LLM prompt rewriter=false2026.03 | 55 | 98 | 74 | 39 | 85 | 15 | — | 23 | — | — | — | — | |
| SDXL2026.04 | 55 | 98 | 74 | 39 | — | 15 | 23 | — | 85 | — | — | — | |
| SD-XLModel Type=Diffusion2026.05 | 55 | 98 | 74 | 39 | 85 | 15 | — | 23 | — | — | — | — | |
| Emu3Model Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 54 | 98 | 71 | 34 | 81 | 17 | — | 21 | — | — | — | — | |
| Emu3-Gen2026.04 | 54 | 98 | 71 | 34 | — | 17 | 21 | — | 81 | — | — | — | |
| Emu3-GenModel Type=Autoregressive2026.05 | 54 | 98 | 71 | 34 | 81 | 17 | — | 21 | — | — | — | — | |
| Show-oModel Type=Generation & Understanding, LLM prompt rewriter=false2026.03 | 53 | 95 | 52 | 49 | 82 | 11 | — | 28 | — | — | — | — | |
| Show-oModel Type=Autoregressive2026.05 | 53 | 95 | 52 | 49 | 82 | 11 | — | 28 | — | — | — | — | |
| HyperAlignBackbone=SD v1.52026.01 | 52 | 98 | 66 | 40 | 82 | 9 | 24 | — | — | — | — | — | |
| DALL-E22026.04 | 52 | 94 | 66 | 49 | — | 10 | 19 | — | 77 | — | — | — | |
| DALLE-2Model Type=Diffusion2026.05 | 52 | 94 | 66 | 49 | 77 | 10 | — | 19 | — | — | — | — | |
| Lumina-mGPT (7B) + SJD++Base Model=Lumina-mGPT (7B), Decoding=SJD with token reuse2025.12 | 51.24 | 98.75 | 62.63 | 36.25 | 79.79 | 10 | 20 | — | — | 27.82 | 3.13 | 6.44 | |
| Emu3-Gen (8B) + SJD++Base Model=Emu3-Gen (8B), Decoding=SJD with token reuse2025.12 | 50.92 | 93.75 | 55.56 | 23.75 | 81.91 | 11 | 13 | — | — | 124.21 | 3.03 | 7.49 | |
| DyMOBackbone=SD v1.52026.01 | 50 | 98 | 60 | 55 | 79 | 5 | 14 | — | — | — | — | — | |
| SD v2.1Model Type=Only Generation, LLM prompt rewriter=false2026.03 | 50 | 98 | 51 | 44 | 85 | 7 | — | 17 | — | — | — | — | |
| SDv2.12026.04 | 50 | 98 | 51 | 44 | — | 7 | 17 | — | 85 | — | — | — | |
| SD2.1Model Type=Diffusion2026.05 | 50 | 98 | 51 | 44 | 85 | 7 | — | 17 | — | — | — | — | |
| ε-greedyBackbone=SD v1.52026.01 | 48 | 98 | 66 | 38 | 72 | 5 | 5 | — | — | — | — | — | |
| Lumina-mGPT (7B)Base Model=Lumina-mGPT (7B), Decoding=Autoregressive2025.12 | 48 | 98.41 | 61.84 | 35.81 | 79.24 | 10.37 | 19.36 | — | — | 87.19 | 1 | 1 | |
| Emu3-Gen (8B)Base Model=Emu3-Gen (8B), Decoding=Autoregressive2025.12 | 48 | 94.1 | 55.18 | 23.42 | 81.45 | 11.37 | 13.28 | — | — | 376.01 | 1 | 1 | |
| Emu3-Gen (8B) + SJDBase Model=Emu3-Gen (8B), Decoding=Speculative Jacobi Decoding2025.12 | 48 | 93.96 | 55.37 | 23.61 | 81.63 | 11.15 | 13.19 | — | — | 207.27 | 1.81 | 2.01 | |
| PixArt-α2026.04 | 48 | 98 | 50 | 44 | — | 8 | 7 | — | 0 | — | — | — | |
| SDXL-TurboSteps=4 Steps2025.12 | 47.66 | 97.81 | 40.4 | 45.62 | — | 11 | — | 14 | 77.13 | — | — | — | |
| Lumina-mGPT (7B) + SJDBase Model=Lumina-mGPT (7B), Decoding=Speculative Jacobi Decoding2025.12 | 47 | 98.56 | 62.11 | 35.94 | 79.42 | 9.81 | 19.52 | — | — | 42.76 | 2.04 | 2.23 | |
| SDXL-TurboSteps=8 Steps2025.12 | 46.52 | 96.88 | 37.63 | 45.62 | — | 11 | — | 13 | 75 | — | — | — | |
| SDXL-TurboSteps=2 Steps2025.12 | 46.22 | 97.81 | 33.08 | 45.94 | — | 10.25 | — | 15 | 75.27 | — | — | — | |
| SDXLSteps=50 Steps2025.12 | 46.01 | 96.88 | 42.17 | 33.12 | — | 9.5 | — | 13 | 81.38 | — | — | — | |
| FreeDoMBackbone=SD v1.52026.01 | 45 | 95 | 49 | 44 | 79 | 6 | 7 | — | — | — | — | — |