Image Reconstruction on MS-COCO 256x256 2017 (val)
32.48PSNRUniFlow(InternViT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| UniFlow(InternViT)Type=Continuous-Diffusion, Training Data=IN-1K, Ratio=142025.10 | 32.48 | 95 | 1.88 | — | |
| FLUX-VAEType=Continuous-Pixel, Training Data=–, Ratio=82025.10 | 32.32 | 93 | 1.35 | — | |
| Qwen-ImageType=Continuous-Pixel, Training Data=–, Ratio=82025.10 | 32.01 | 91 | 4.62 | — | |
| Wan2.1Type=Continuous-Pixel, Training Data=–, Ratio=82025.10 | 31.19 | 90 | 3.45 | — | |
| SD-VAE 3Type=Continuous-Pixel, Training Data=–, Ratio=82025.10 | 31.18 | 89 | 1.67 | — | |
| Wan2.2Type=Continuous-Pixel, Training Data=–, Ratio=162025.10 | 31.1 | 89 | 3.28 | — | |
| UniFlow(DINOv2)Type=Continuous-Diffusion, Training Data=IN-1K, Ratio=142025.10 | 30.66 | 94 | 2.81 | — | |
| UniFlow(CLIP)Type=Continuous-Diffusion, Training Data=IN-1K, Ratio=142025.10 | 29.61 | 92 | 3.69 | — | |
| VA-VAEType=Continuous-Pixel, Training Data=1N-1K, Ratio=162025.10 | 27.5 | 81 | 2.71 | — | |
| SD-VAE XLType=Continuous-Pixel, Training Data=OImg+LAae++, Ratio=82025.10 | 27.08 | 80 | 3.93 | — | |
| SD-VAE 2.xType=Continuous-Pixel, Training Data=OImg+LAae, Ratio=82025.10 | 26.62 | 77 | 4.26 | — | |
| OmniTokenizerType=Continuous-Pixel, Training Data=IN-1K+K600, Ratio=82025.10 | 26.44 | 83 | 4.69 | — | |
| UniFlow(SigLIP2)Type=Continuous-Diffusion, Training Data=IN-1K, Ratio=162025.10 | 26.38 | 86 | 3.44 | — | |
| Cosmos-CIType=Continuous-Pixel, Training Data=–, Ratio=162025.10 | 24.74 | 71 | 5.06 | — | |
| VibeToken-LL/256Type=1D2026.04 | 24.71 | 82.8 | — | 3.59 | |
| SD-VAE 1.xType=Continuous-Pixel, Training Data=OImg, Ratio=82025.10 | 23.21 | 69 | 5.94 | — | |
| IBQType=2D2026.04 | 22.56 | 60.3 | — | 6.14 | |
| Open-MAGVIT2Type=Discrete-Pixel, Training Data=Mixed100M, Ratio=162025.10 | 22.31 | 65 | 6.76 | — | |
| TiTok-SL/256Type=1D2026.04 | 21.6 | 72.6 | — | 6.84 | |
| Show-oType=Discrete-Pixel, Training Data=-, Ratio=162025.10 | 20.9 | 59 | 9.26 | — | |
| LlamaGenType=Discrete-Pixel, Training Data=MS+IN-1K, Ratio=162025.10 | 20.28 | 55 | 8.4 | — | |
| Cosmos-DIType=Discrete-Pixel, Training Data=–, Ratio=162025.10 | 19.22 | 48 | 11.97 | — | |
| One-D-Piece-LL/256Type=1D2026.04 | 18.41 | 61.2 | — | 8.02 | |
| SeTokType=Discrete-Pixel, Training Data=IN-1K+OImg, Ratio=–2025.10 | — | — | 2.07 | — |