Text-to-Image Generation on COCO 256 x 256 2014 (val)
3.22FIDParti
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PartiType=Autoregressive, Training datasets=LAION (400M) + FIT (400M) + JFT (4B), #Params=20B + 630M (AE), Setting=Trained on external large dataset with access to MS-COCO, Finetuned on MS-COCO=true2022.09 | 3.22 | — | — | |
| Re-Imagen+Type=Diffusion, Training datasets=KNN-ImageText (50M), #Params=2.5B + 750M (SR), Setting=Trained on external large dataset with access to MS-COCO2022.09 | 5.25 | — | — | |
| U-ViT-S/2 (Deep)Type=Latent diffusion, Training datasets=MS-COCO (83K), #Params=58M + 123M (TE) + 84M (AE), Setting=Trained on MS-COCO2022.09 | 5.48 | — | — | |
| U-ViT-S/2Type=Latent diffusion, Training datasets=MS-COCO (83K), #Params=45M + 123M (TE) + 84M (AE), Setting=Trained on MS-COCO2022.09 | 5.95 | — | — | |
| Re-ImagenType=Diffusion, Training datasets=KNN-ImageText (50M), #Params=2.5B + 750M (SR), Setting=Zero-shot on external large dataset2022.09 | 6.88 | — | — | |
| Re-ImagenType=Diffusion, Training datasets=KNN-ImageText (50M), #Params=2.5B + 750M (SR), Zero-shot=true2025.12 | 6.88 | — | — | |
| FAE (DinoV2)+T5Type=Diffusion, Training datasets=CC12M, #Params=604M+ 514M (FAE), Zero-shot=true, Classifier-Free Guidance (CFG)=true2025.12 | 6.9 | — | — | |
| FAE (SigLIPV2)+T5Type=Diffusion, Training datasets=CC12M, #Params=604M+ 514M (FAE), Zero-shot=true, Classifier-Free Guidance (CFG)=true2025.12 | 7.11 | — | — | |
| PartiType=Autoregressive, Training datasets=LAION (400M) + FIT (400M) + JFT (4B), #Params=20B + 630M (AE), Setting=Zero-shot on external large dataset2022.09 | 7.23 | — | — | |
| PartiType=Autoregressive, Training datasets=LAION (400M) + FIT (400M) + JFT (4B), #Params=20B + 630M (AE), Zero-shot=true2025.12 | 7.23 | — | — | |
| ImagenType=Diffusion, Training datasets=Internal dataset (460M) + LAION (400M), #Params=2B + 4.6B (TE) + 600M (SR), Setting=Zero-shot on external large dataset2022.09 | 7.27 | — | — | |
| ImagenType=Diffusion, Training datasets=Internal dataset (460M) + LAION (400M), #Params=2B + 4.6B (TE) + 600M (SR), Zero-shot=true2025.12 | 7.27 | — | — | |
| U-Net*Type=Latent diffusion, Training datasets=MS-COCO (83K), #Params=53M + 123M (TE) + 84M (AE), Setting=Trained on MS-COCO, Retrieved from knowledge base=MS-COCO2022.09 | 7.32 | — | — | |
| FAE (DinoV2)+T5Type=Diffusion, Training datasets=CC12M, #Params=604M+ 514M (FAE), Zero-shot=true, Classifier-Free Guidance (CFG)=false2025.12 | 7.47 | — | — | |
| Make-A-SceneType=Autoregressive, Training datasets=Union datasets (with MS-COCO) (35M), #Params=4B, Setting=Trained on external large dataset with access to MS-COCO2022.09 | 7.55 | — | — | |
| FAE (SigLIPV2)+T5Type=Diffusion, Training datasets=CC12M, #Params=604M+ 514M (FAE), Zero-shot=true, Classifier-Free Guidance (CFG)=false2025.12 | 7.57 | — | — | |
| LAFITEType=GAN, Training datasets=MS-COCO (83K), #Params=75M + 151M (TE), Setting=Trained on MS-COCO2022.09 | 8.12 | — | — | |
| FriroType=Diffusion, Training datasets=MS-COCO (83K), #Params=512M + 186M (TE) + 68M (AE), Setting=Trained on MS-COCO2022.09 | 8.97 | — | — | |
| XMC-GANType=GAN, Training datasets=MS-COCO (83K), #Params=166M, Setting=Trained on MS-COCO2022.09 | 9.33 | — | — | |
| DALL-E 2Type=Diffusion, Training datasets=DALL-E dataset (250M), #Params=4.5B + 700M (SR), Setting=Zero-shot on external large dataset2022.09 | 10.39 | — | — | |
| DALL-E 2Type=Diffusion, Training datasets=DALL-E dataset (250M), #Params=4.5B + 700M (SR), Zero-shot=true2025.12 | 10.39 | — | — | |
| Make-A-SceneType=Autoregressive, Training datasets=Union datasets (without MS-COCO) (35M), #Params=4B, Setting=Zero-shot on external large dataset2022.09 | 11.84 | — | — | |
| Make-A-SceneType=Autoregressive, Training datasets=Union datasets (without MS-COCO) (35M), #Params=4B, Zero-shot=true2025.12 | 11.84 | — | — | |
| GLIDEType=Diffusion, Training datasets=DALL-E dataset (250M), #Params=3.5B + 1.5B (SR), Setting=Zero-shot on external large dataset2022.09 | 12.24 | — | — | |
| GLIDEType=Diffusion, Training datasets=DALL-E dataset (250M), #Params=3.5B + 1.5B (SR), Zero-shot=true2025.12 | 12.24 | — | — | |
| VQ-Diffusion+Type=Discrete diffusion, Training datasets=Conceptual Caption Subset (7M), #Params=370M, Setting=Trained on external large dataset with access to MS-COCO2022.09 | 13.86 | — | — | |
| DF-GANType=GAN, Training datasets=MS-COCO (83K), #Params=19M, Setting=Trained on MS-COCO2022.09 | 19.32 | — | — | |
| VQ-DiffusionType=Discrete diffusion, Training datasets=MS-COCO (83K), #Params=370M, Setting=Trained on MS-COCO2022.09 | 19.75 | — | — | |
| SDVAE+T5Type=Diffusion, Training datasets=CC12M, #Params=604M, Zero-shot=true, Classifier-Free Guidance (CFG)=false2025.12 | 21.25 | — | — | |
| LAFITEType=GAN, Training datasets=CC3M (3M), #Params=75M + 151M (TE), Setting=Zero-shot on external large dataset2022.09 | 26.94 | — | — | |
| LAFITEType=GAN, Training datasets=CC3M (3M), #Params=75M + 151M (TE), Zero-shot=true2025.12 | 26.94 | — | — | |
| CogViewType=Autoregressive, Training datasets=Internal dataset (30M), #Params=4B, Setting=Zero-shot on external large dataset2022.09 | 27.1 | — | — | |
| CogViewType=Autoregressive, Training datasets=Internal dataset (30M), #Params=4B, Zero-shot=true2025.12 | 27.1 | — | — | |
| DALL-EType=Autoregressive, Training datasets=DALL-E dataset (250M), #Params=12B, Setting=Zero-shot on external large dataset2022.09 | 28 | — | — | |
| DALL-EType=Autoregressive, Training datasets=DALL-E dataset (250M), #Params=12B, Zero-shot=true2025.12 | 28 | — | — | |
| DM-GANType=GAN, Training datasets=MS-COCO (83K), #Params=46M, Setting=Trained on MS-COCO2022.09 | 32.64 | — | — | |
| AttnGANType=GAN, Training datasets=MS-COCO (83K), #Params=230M, Setting=Trained on MS-COCO2022.09 | 35.49 | — | — | |
| OriginalSource=Reference COCO Images2022.05 | — | 50 | — |