Text-to-Image Generation on T2I-CompBench (test)
91Color AccuracyRewardFlow
Evaluation results
| Method | Links | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RewardFlowBase Model=Qwen Image2026.04 | 91 | — | — | — | — | 83 | 90 | 39 | 51 | 78 | — | — | — | — | — | — | — | — | |
| RewardFlowBase Model=Flux2026.04 | 88 | — | — | — | — | 69 | 78 | 33 | 42 | 57 | — | — | — | — | — | — | — | — | |
| ReNOBase Model=Qwen Image2026.04 | 84 | — | — | — | — | 75 | 84 | 36 | 43 | 63 | — | — | — | — | — | — | — | — | |
| Stable Diffusion 3.5-medium + MarmotModel=Stable Diffusion 3.5-medium, Enhancement=Marmot2025.04 | 83.2 | — | — | — | — | 61.28 | 73.79 | 38.03 | 31.66 | 37.37 | — | — | — | — | — | — | — | — | |
| Qwen ImageBase Model=Qwen Image2026.04 | 83 | — | — | — | — | 72 | 80 | 35 | 39 | 61 | — | — | — | — | — | — | — | — | |
| DALL-E 32024.06 | 81 | — | — | — | — | 68 | 81 | — | — | — | — | — | — | — | — | — | — | — | |
| DALL-E 32026.04 | 81 | — | — | — | — | 68 | 81 | — | — | — | — | — | — | — | — | — | — | — | |
| ReNOBase Model=Flux2026.04 | 81 | — | — | — | — | 64 | 72 | 29 | 35 | 49 | — | — | — | — | — | — | — | — | |
| Stable Diffusion 3.5-mediumModel=Stable Diffusion 3.5-medium2025.04 | 79.85 | — | — | — | — | 57.7 | 72 | 29.89 | 31.49 | 36.37 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=HyperSDXL2024.06 | 79.04 | — | — | — | — | 63.24 | 76.71 | 26.16 | 31.45 | 47.66 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=HyperSDXL2024.06 | 79 | — | — | — | — | 63 | 77 | 26 | 31 | 48 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=SD-Turbo2024.06 | 78.3 | — | — | — | — | 62.44 | 74.66 | 22.35 | 31.61 | 48.29 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=SDXL-Turbo2024.06 | 78 | — | — | — | — | 59.55 | 73.96 | 25.51 | 31.47 | 46.9 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=SD-Turbo2024.06 | 78 | — | — | — | — | 62 | 75 | 22 | 32 | 48 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=SDXL-Turbo2024.06 | 78 | — | — | — | — | 60 | 74 | 26 | 31 | 47 | — | — | — | — | — | — | — | — | |
| FLUX.1-dev + MarmotModel=FLUX.1-dev, Enhancement=Marmot2025.04 | 77.76 | — | — | — | — | 52.47 | 67.35 | 34.3 | 30.85 | 37.69 | — | — | — | — | — | — | — | — | |
| GenTronText Encoder=CLIP-T5XXL, Model Scale=G/22023.12 | 76.74 | — | — | 49.99 | — | 57 | 71.5 | 20.98 | 32.02 | 41.67 | — | — | — | — | — | — | — | — | |
| SD-v1.5Params.=0.9 B, Data=2 B2025.03 | 76 | 97 | 38 | 43 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FluxBase Model=Flux2026.04 | 75 | — | — | — | — | 61 | 69 | 26 | 33 | 47 | — | — | — | — | — | — | — | — | |
| FLUX.1-devModel=FLUX.1-dev2025.04 | 74.79 | — | — | — | — | 46.45 | 62.11 | 27.9 | 30.67 | 36.49 | — | — | — | — | — | — | — | — | |
| RewardFlowBase Model=PixArt-α DMD2026.04 | 74 | — | — | — | — | 66 | 75 | 30 | 39 | 52 | — | — | — | — | — | — | — | — | |
| ImageSelectBackbone=SD-Turbo2024.06 | 72.22 | — | — | — | — | 55.52 | 69.19 | 22.16 | 31.54 | 46.18 | — | — | — | — | — | — | — | — | |
| PixArt-α2024.06 | 69 | — | — | — | — | 56 | 70 | 21 | 32 | 41 | — | — | — | — | — | — | — | — | |
| PixArt-α2026.04 | 69 | — | — | — | — | 56 | 70 | 21 | 32 | 41 | — | — | — | — | — | — | — | — | |
| PIXART-α2023.09 | 68.86 | — | — | — | — | 55.82 | 70.44 | 20.82 | 31.79 | 41.17 | — | — | — | — | — | — | — | — | |
| PixArt-α2023.12 | 68.86 | — | — | 48.15 | — | 55.82 | 70.44 | 20.82 | 31.79 | 41.17 | — | — | — | — | — | — | — | — | |
| ARPG-XLParams.=0.8 B, Data=4 M2025.03 | 68 | 87 | 30 | 31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GORS2023.09 | 66.03 | — | — | — | — | 47.85 | 62.87 | 18.15 | 31.93 | 33.28 | — | — | — | — | — | — | — | — | |
| GORS2023.12 | 66.03 | — | — | 43.35 | — | 47.85 | 62.87 | 18.15 | 31.93 | 33.28 | — | — | — | — | — | — | — | — | |
| GORSBackbone=SD2.12024.06 | 66.03 | — | — | — | — | 47.85 | 62.87 | 18.15 | 31.93 | 33.28 | — | — | — | — | — | — | — | — | |
| Emu3-Gen (8B)Model=Emu3-Gen (8B), Decoding Strategy=Autoregressive2025.12 | 65.49 | — | — | — | — | 36.63 | 50.61 | — | — | — | — | 374.89 | 1 | 1 | 20.66 | 31.05 | 31.08 | 36.07 | |
| HyperSDXL2024.06 | 65.35 | — | — | — | — | 49.56 | 64.96 | 25.09 | 31.08 | 45.82 | — | — | — | — | — | — | — | — | |
| Emu3-Gen (8B) + SJDModel=Emu3-Gen (8B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 65.25 | — | — | — | — | 36.28 | 50.46 | — | — | — | — | 206.59 | 1.81 | 2.02 | 20.68 | 30.89 | 30.97 | 35.95 | |
| Emu3-Gen (8B) + SJD++Model=Emu3-Gen (8B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 65.08 | — | — | — | — | 36.09 | 50.29 | — | — | — | — | 124.52 | 3.01 | 7.48 | 20.71 | 30.69 | 30.85 | 35.83 | |
| HyperSDXL2024.06 | 65 | — | — | — | — | 50 | 65 | 25 | 31 | 46 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=PixArt-a DMD2024.06 | 64.54 | — | — | — | — | 56.58 | 71.86 | 25.08 | 31.38 | 45.54 | — | — | — | — | — | — | — | — | |
| SimpleAR (0.5B)Model=SimpleAR (0.5B), Decoding Strategy=Autoregressive2025.12 | 64.02 | — | — | — | — | 35.23 | 49.56 | — | — | — | — | 83.61 | 1 | 1 | 20.58 | 30.6 | 30.9 | 35.58 | |
| LlamaGen-XLParams.=0.8 B, Data=60 M2025.03 | 64 | 87 | 34 | 32 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Attn-Exct v22023.09 | 64 | — | — | — | — | 45.17 | 59.63 | 14.55 | 31.09 | 34.01 | — | — | — | — | — | — | — | — | |
| Attn-Exct v22023.12 | 64 | — | — | 41.41 | — | 45.17 | 59.63 | 14.55 | 31.09 | 34.01 | — | — | — | — | — | — | — | — | |
| Attn-ExctBackbone=SD2.12024.06 | 64 | — | — | — | — | 45.17 | 59.63 | 14.55 | 31.09 | 34.01 | — | — | — | — | — | — | — | — | |
| SDXL2024.06 | 64 | — | — | — | — | 54 | 56 | 20 | 31 | 41 | — | — | — | — | — | — | — | — | |
| ReNOBackbone=PixArt-α DMD2024.06 | 64 | — | — | — | — | 57 | 72 | 25 | 31 | 46 | — | — | — | — | — | — | — | — | |
| SDXL2026.04 | 64 | — | — | — | — | 54 | 56 | 20 | 31 | 41 | — | — | — | — | — | — | — | — | |
| ReNOBase Model=PixArt-α DMD2026.04 | 64 | — | — | — | — | 57 | 72 | 25 | 31 | 46 | — | — | — | — | — | — | — | — | |
| SimpleAR (0.5B) + SJDModel=SimpleAR (0.5B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 63.74 | — | — | — | — | 35.05 | 49.3 | — | — | — | — | 42.93 | 1.95 | 2.31 | 20.45 | 30.38 | 30.74 | 35.41 | |
| SDXL2023.09 | 63.69 | — | — | — | — | 54.08 | 56.37 | 20.32 | 31.1 | 40.91 | — | — | — | — | — | — | — | — | |
| LDM XL2023.12 | 63.69 | — | — | 44.41 | — | 54.08 | 56.37 | 20.32 | 31.1 | 40.91 | — | — | — | — | — | — | — | — | |
| Janus-Pro (1B)Model=Janus-Pro (1B), Decoding Strategy=Autoregressive2025.12 | 63.65 | — | — | — | — | 35.32 | 49.46 | — | — | — | — | 14.4 | 1 | 1 | 20.69 | 30.76 | 30.94 | 35.66 | |
| SimpleAR (0.5B) + SJD++Model=SimpleAR (0.5B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 63.46 | — | — | — | — | 34.81 | 49.11 | — | — | — | — | 29.82 | 2.81 | 3.33 | 20.33 | 30.21 | 30.61 | 35.27 | |
| Janus-Pro (1B) + SJDModel=Janus-Pro (1B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 63.28 | — | — | — | — | 35.04 | 49.24 | — | — | — | — | 8.03 | 1.8 | 1.86 | 20.59 | 30.43 | 30.72 | 35.46 | |
| Janus-Pro (1B) + SJD++Model=Janus-Pro (1B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 62.97 | — | — | — | — | 34.84 | 48.99 | — | — | — | — | 6.32 | 2.28 | 2.47 | 20.5 | 30.1 | 30.52 | 35.3 | |
| SDXL-Turbo2024.06 | 61.49 | — | — | — | — | 43.66 | 60.01 | 24.01 | 31.18 | 42.5 | — | — | — | — | — | — | — | — | |
| SDXL-Turbo2024.06 | 61 | — | — | — | — | 44 | 60 | 24 | 31 | 43 | — | — | — | — | — | — | — | — | |
| LlamaGen (0.8B)Model=LlamaGen (0.8B), Decoding Strategy=Autoregressive2025.12 | 60.14 | — | — | — | — | 34.02 | 45.61 | — | — | — | — | 44.92 | 1 | 1 | 20.37 | 29.55 | 28.93 | 33.98 | |
| LlamaGen (0.8B) + SJDModel=LlamaGen (0.8B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 59.89 | — | — | — | — | 33.84 | 45.39 | — | — | — | — | 32.1 | 1.4 | 2.11 | 20.29 | 29.36 | 28.78 | 33.84 | |
| LlamaGen (0.8B) + SJD++Model=LlamaGen (0.8B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 59.65 | — | — | — | — | 33.68 | 45.21 | — | — | — | — | 19.56 | 2.3 | 3.23 | 20.2 | 29.19 | 28.62 | 33.7 | |
| Dalle-22023.09 | 57.5 | — | — | — | — | 54.64 | 63.74 | 12.83 | 30.43 | 36.96 | — | — | — | — | — | — | — | — | |
| DALL-E 22023.12 | 57.5 | — | — | 42.68 | — | 54.64 | 63.74 | 12.83 | 30.43 | 36.96 | — | — | — | — | — | — | — | — | |
| DALL-E 22024.06 | 57 | — | — | — | — | 55 | 64 | 13 | 30 | 37 | — | — | — | — | — | — | — | — | |
| DALL-E 22026.04 | 57 | — | — | — | — | 55 | 64 | 13 | 30 | 37 | — | — | — | — | — | — | — | — | |
| Lumina-mGPT (7B)Model=Lumina-mGPT (7B), Decoding Strategy=Autoregressive2025.12 | 56.28 | — | — | — | — | 33.04 | 41.29 | — | — | — | — | 87.21 | 1 | 1 | 20.06 | 28.37 | 26.1 | 31.79 | |
| Lumina-mGPT (7B) + SJDModel=Lumina-mGPT (7B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 56.03 | — | — | — | — | 32.93 | 41.09 | — | — | — | — | 42.88 | 2.04 | 2.23 | — | 28.21 | 25.93 | 31.65 | |
| Lumina-mGPT (7B) + SJD++Model=Lumina-mGPT (7B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 55.78 | — | — | — | — | 32.79 | 41.07 | — | — | — | — | 27.97 | 3.13 | 6.42 | 19.96 | 27.99 | 25.82 | 31.63 | |
| SD-Turbo2024.06 | 55.13 | — | — | — | — | 44.48 | 56.9 | 17.39 | 31.01 | 40.52 | — | — | — | — | — | — | — | — | |
| SD-Turbo2024.06 | 55 | — | — | — | — | 44 | 57 | 17 | 31 | 41 | — | — | — | — | — | — | — | — | |
| SD v2.12024.06 | 51 | — | — | — | — | 42 | 49 | 13 | 31 | 34 | — | — | — | — | — | — | — | — | |
| SD v2.12026.04 | 51 | — | — | — | — | 42 | 49 | 13 | 31 | 34 | — | — | — | — | — | — | — | — | |
| Stable v22023.09 | 50.65 | — | — | — | — | 42.21 | 49.22 | 13.42 | 30.96 | 33.86 | — | — | — | — | — | — | — | — | |
| LDM v22023.12 | 50.65 | — | — | 36.72 | — | 42.21 | 49.22 | 13.42 | 30.96 | 33.86 | — | — | — | — | — | — | — | — | |
| SD2.12024.06 | 50.65 | — | — | — | — | 42.21 | 49.22 | 13.42 | 30.96 | 33.86 | — | — | — | — | — | — | — | — | |
| TokenComposeBackbone=SD2.12024.06 | 50.55 | — | — | — | — | 48.52 | 58.81 | 18.15 | 31.73 | 29.37 | — | — | — | — | — | — | — | — | |
| Structured v22023.09 | 49.9 | — | — | — | — | 42.18 | 49 | 13.86 | 31.11 | 33.55 | — | — | — | — | — | — | — | — | |
| Structured v22023.12 | 49.9 | — | — | 36.6 | — | 42.18 | 49 | 13.86 | 31.11 | 33.55 | — | — | — | — | — | — | — | — | |
| StructureDiffusionBackbone=SD2.12024.06 | 49.9 | — | — | — | — | 42.18 | 49 | 13.86 | 31.11 | 33.55 | — | — | — | — | — | — | — | — | |
| Attn-Mask-ControlBackbone=SD2.12024.06 | 41.19 | — | — | — | — | 46.49 | 45.05 | 12.49 | 30.46 | 37.79 | — | — | — | — | — | — | — | — | |
| Composable v22023.09 | 40.63 | — | — | — | — | 32.99 | 36.45 | 8 | 29.8 | 28.98 | — | — | — | — | — | — | — | — | |
| Composable v22023.12 | 40.63 | — | — | 29.47 | — | 32.99 | 36.45 | 8 | 29.8 | 28.98 | — | — | — | — | — | — | — | — | |
| Composable DiffusionBackbone=SD2.12024.06 | 40.63 | — | — | — | — | 32.99 | 36.45 | 8 | 29.8 | 28.98 | — | — | — | — | — | — | — | — | |
| PixArt-a DMD2024.06 | 38.24 | — | — | — | — | 34.14 | 46.91 | 19.06 | 30.6 | 36.43 | — | — | — | — | — | — | — | — | |
| SD v1.42024.06 | 38 | — | — | — | — | 36 | 42 | 12 | 31 | 31 | — | — | — | — | — | — | — | — | |
| PixArt-α DMD2024.06 | 38 | — | — | — | — | 34 | 47 | 19 | 30 | 36 | — | — | — | — | — | — | — | — | |
| SD v1.42026.04 | 38 | — | — | — | — | 36 | 42 | 12 | 31 | 31 | — | — | — | — | — | — | — | — | |
| PixArt-α DMDBase Model=PixArt-α DMD2026.04 | 38 | — | — | — | — | 34 | 47 | 19 | 30 | 36 | — | — | — | — | — | — | — | — | |
| Stable v1.42023.09 | 37.65 | — | — | — | — | 35.76 | 41.56 | 12.46 | 30.79 | 30.8 | — | — | — | — | — | — | — | — | |
| LDM v1.42023.12 | 37.65 | — | — | 31.5 | — | 35.76 | 41.56 | 12.46 | 30.79 | 30.8 | — | — | — | — | — | — | — | — | |
| ChameleonParams.=7.0 B, Data=1.4 B2025.03 | — | — | — | 39 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Flux.1-devArch.=DiT, Param.=12B, FPS=0.04, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 47.1 | — | — | — | — | — | — | — | |
| IF-XLArch.=U-Net, Param.=5.5B, FPS=0.06, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 42.1 | — | — | — | — | — | — | — | |
| LUMINA-NextArch.=DiT, Param.=2.0B, FPS=0.06, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 35.3 | — | — | — | — | — | — | — | |
| Ours-fullArch.=DiT, Param.=1.6B, FPS=0.28, Latency=15802026.01 | — | — | — | — | — | — | — | — | — | — | 53.6 | — | — | — | — | — | — | — | |
| Ours-smallArch.=DiT, Param.=0.4B, FPS=0.62, Latency=3602026.01 | — | — | — | — | — | — | — | — | — | — | 50.6 | — | — | — | — | — | — | — | |
| Ours-tinyArch.=DiT, Param.=0.3B, FPS=0.81, Latency=2802026.01 | — | — | — | — | — | — | — | — | — | — | 50.2 | — | — | — | — | — | — | — | |
| PixArt-αArch.=DiT, Param.=0.6B, FPS=0.42, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 35.1 | — | — | — | — | — | — | — | |
| PixArt-ΣArch.=DiT, Param.=0.6B, FPS=0.46, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 42.7 | — | — | — | — | — | — | — | |
| Playgroundv2.5Arch.=DiT, Param.=2.6B, FPS=0.18, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 23.7 | — | — | — | — | — | — | — | |
| SANAArch.=Hybrid, Param.=1.6B, FPS=0.91, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 47.6 | — | — | — | — | — | — | — | |
| SD3-MediumArch.=DiT, Param.=2.0B, FPS=0.28, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 52.2 | — | — | — | — | — | — | — | |
| SD3.5-LargeArch.=DiT, Param.=8.1B, FPS=0.08, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 50.7 | — | — | — | — | — | — | — | |
| SDXLArch.=U-Net, Param.=2.6B, FPS=0.18, Latency=OOM2026.01 | — | — | — | — | — | — | — | — | — | — | 40.2 | — | — | — | — | — | — | — |