{"url":"/sota/image-generation-on-celeba-hq-256x256","task":{"name":"Image Generation","url":"/task/image-generation","note":null},"dataset":{"name":"CelebA-HQ 256x256","url":"/dataset/celeba-hq"},"category":"Computer Vision","categories":["Computer Vision","Medical","Miscellaneous","Natural Language Processing"],"category_note":null,"description":"**Image Generation** (synthesis) is the task of generating new images from an existing dataset.\r\n\r\n- **Unconditional generation** refers to generating samples unconditionally from the dataset, i.e. $p(y)$\r\n- **[Conditional image generation](/task/conditional-image-generation)** (subtask) refers to generating samples conditionally from the dataset, based on a label, i.e. $p(y|x)$.\r\n\r\nIn this section, you can find state-of-the-art leaderboards for **unconditional generation**. For conditional  generation, and other types of image generations, refer to the subtasks.\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [StyleGAN](https://github.com/NVlabs/stylegan) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["FID","bits/dimension","Recall","NFE","Precision","clean-FID"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"FID":"lower","bits/dimension":null,"Recall":"higher","NFE":null,"Precision":"higher","clean-FID":"lower"}},"counts":{"rows":19,"rows_with_code":15,"rows_with_paper_page":19,"rows_dated":19,"rows_using_additional_data":1},"rows":[{"rank_in_archive_order":1,"model":"RDM","metrics":{"FID":"3.15","Precision":"0.77","Recall":"0.55"},"uses_additional_data":true,"paper_date":"2023-09-04","paper":"/paper/relay-diffusion-unifying-diffusion-process-1","paper_url":"https://arxiv.org/abs/2309.03350v1","paper_title":"Relay Diffusion: Unifying diffusion process across resolutions for image synthesis","code":"https://github.com/THUDM/RelayDiffusion","n_code_links":1,"syntology":{"n_ran":14,"n_unverified":3,"n_samples":17,"n_pointer_only_licence":0}},{"rank_in_archive_order":2,"model":"StyleSwin","metrics":{"FID":"3.25"},"uses_additional_data":false,"paper_date":"2021-12-20","paper":"/paper/styleswin-transformer-based-gan-for-high-1","paper_url":"https://arxiv.org/abs/2112.10762v2","paper_title":"StyleSwin: Transformer-based GAN for High-resolution Image Generation","code":"https://github.com/microsoft/StyleSwin","n_code_links":1,"syntology":{"n_ran":5,"n_unverified":0,"n_samples":5,"n_pointer_only_licence":3}},{"rank_in_archive_order":3,"model":"LDM-4","metrics":{"FID":"5.11"},"uses_additional_data":false,"paper_date":"2021-12-20","paper":"/paper/high-resolution-image-synthesis-with-latent","paper_url":"https://arxiv.org/abs/2112.10752v2","paper_title":"High-Resolution Image Synthesis with Latent Diffusion Models","code":"https://github.com/compvis/stable-diffusion","n_code_links":41,"syntology":{"n_ran":21,"n_unverified":7,"n_samples":28,"n_pointer_only_licence":5}},{"rank_in_archive_order":4,"model":"LFM","metrics":{"FID":"5.26"},"uses_additional_data":false,"paper_date":"2023-07-17","paper":"/paper/flow-matching-in-latent-space","paper_url":"https://arxiv.org/abs/2307.08698v1","paper_title":"Flow Matching in Latent Space","code":"https://github.com/vinairesearch/lfm","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":5,"model":"RDUOT","metrics":{"FID":"5.6","Recall":"0.38"},"uses_additional_data":false,"paper_date":"2023-11-28","paper":"/paper/robust-diffusion-gan-using-semi-unbalanced","paper_url":"https://arxiv.org/abs/2311.17101v2","paper_title":"A High-Quality Robust Diffusion Framework for Corrupted Dataset","code":"https://github.com/VinAIResearch/RDUOT","n_code_links":1,"syntology":{"n_ran":5,"n_unverified":7,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":6,"model":"WaveDiff","metrics":{"FID":"5.94","NFE":"2","Recall":"0.37"},"uses_additional_data":false,"paper_date":"2022-11-29","paper":"/paper/wavelet-diffusion-models-are-fast-and","paper_url":"https://arxiv.org/abs/2211.16152v2","paper_title":"Wavelet Diffusion Models are fast and scalable Image Generators","code":"https://github.com/vinairesearch/wavediff","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"UNCSN++ (RVE) + ST","metrics":{"FID":"7.16"},"uses_additional_data":false,"paper_date":"2021-06-10","paper":"/paper/score-matching-model-for-unbounded-data-score-1","paper_url":"https://arxiv.org/abs/2106.05527v5","paper_title":"Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimation","code":"https://github.com/Kim-Dongjun/Soft-Truncation","n_code_links":1,"syntology":{"n_ran":4,"n_unverified":8,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":8,"model":"LSGM","metrics":{"FID":"7.22"},"uses_additional_data":false,"paper_date":"2021-06-10","paper":"/paper/score-based-generative-modeling-in-latent","paper_url":"https://arxiv.org/abs/2106.05931v3","paper_title":"Score-based Generative Modeling in Latent Space","code":"https://github.com/NVlabs/LSGM","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":3,"n_samples":5,"n_pointer_only_licence":5}},{"rank_in_archive_order":9,"model":"DDGAN","metrics":{"FID":"7.64"},"uses_additional_data":false,"paper_date":"2021-12-15","paper":"/paper/tackling-the-generative-learning-trilemma-1","paper_url":"https://arxiv.org/abs/2112.07804v2","paper_title":"Tackling the Generative Learning Trilemma with Denoising Diffusion GANs","code":"https://github.com/NVlabs/denoising-diffusion-gan","n_code_links":5,"syntology":{"n_ran":6,"n_unverified":6,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"DDMI","metrics":{"FID":"8.73"},"uses_additional_data":false,"paper_date":"2024-01-23","paper":"/paper/ddmi-domain-agnostic-latent-diffusion-models","paper_url":"https://arxiv.org/abs/2401.12517v2","paper_title":"DDMI: Domain-Agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations","code":"https://github.com/mlvlab/DDMI","n_code_links":1,"syntology":{"n_ran":17,"n_unverified":4,"n_samples":21,"n_pointer_only_licence":0}},{"rank_in_archive_order":11,"model":"Diffusion-JEBM","metrics":{"FID":"8.78"},"uses_additional_data":false,"paper_date":"2024-05-22","paper":"/paper/learning-latent-space-hierarchical-ebm","paper_url":"https://arxiv.org/abs/2405.13910v2","paper_title":"Learning Latent Space Hierarchical EBM Diffusion Models","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":12,"model":"Joint-EBM","metrics":{"FID":"9.89"},"uses_additional_data":false,"paper_date":"2023-06-10","paper":"/paper/learning-joint-latent-space-ebm-prior-model-1","paper_url":"https://arxiv.org/abs/2306.06323v2","paper_title":"Learning Joint Latent Space EBM Prior Model for Multi-layer Generator","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":13,"model":"VQGAN+Transformer","metrics":{"FID":"10.2"},"uses_additional_data":false,"paper_date":"2020-12-17","paper":"/paper/taming-transformers-for-high-resolution-image","paper_url":"https://arxiv.org/abs/2012.09841v3","paper_title":"Taming Transformers for High-Resolution Image Synthesis","code":"https://github.com/CompVis/taming-transformers","n_code_links":13,"syntology":{"n_ran":6,"n_unverified":0,"n_samples":6,"n_pointer_only_licence":4}},{"rank_in_archive_order":14,"model":"DC-VAE","metrics":{"FID":"15.81"},"uses_additional_data":false,"paper_date":"2020-11-19","paper":"/paper/dual-contradistinctive-generative-autoencoder-1","paper_url":"https://arxiv.org/abs/2011.10063v1","paper_title":"Dual Contradistinctive Generative Autoencoder","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":15,"model":"Dual-MCMC EBM","metrics":{"FID":"15.89"},"uses_additional_data":false,"paper_date":"2023-12-05","paper":"/paper/learning-energy-based-model-via-dual-mcmc-1","paper_url":"https://arxiv.org/abs/2312.02469v1","paper_title":"Learning Energy-based Model via Dual-MCMC Teaching","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":16,"model":"VAEBM","metrics":{"FID":"20.38"},"uses_additional_data":false,"paper_date":"2020-10-01","paper":"/paper/vaebm-a-symbiosis-between-variational","paper_url":"https://arxiv.org/abs/2010.00654v3","paper_title":"VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models","code":"https://github.com/NVlabs/VAEBM","n_code_links":1,"syntology":{"n_ran":6,"n_unverified":3,"n_samples":9,"n_pointer_only_licence":9}},{"rank_in_archive_order":17,"model":"GLOW","metrics":{"FID":"68.93"},"uses_additional_data":false,"paper_date":"2018-07-09","paper":"/paper/glow-generative-flow-with-invertible-1x1","paper_url":"http://arxiv.org/abs/1807.03039v2","paper_title":"Glow: Generative Flow with Invertible 1x1 Convolutions","code":"https://github.com/openai/glow","n_code_links":27,"syntology":{"n_ran":75,"n_unverified":54,"n_samples":129,"n_pointer_only_licence":44}},{"rank_in_archive_order":18,"model":"RNODE","metrics":{"bits/dimension":"1.04"},"uses_additional_data":false,"paper_date":"2020-02-07","paper":"/paper/how-to-train-your-neural-ode","paper_url":"https://arxiv.org/abs/2002.02798v3","paper_title":"How to train your neural ODE: the world of Jacobian and kinetic regularization","code":"https://github.com/cfinlay/ffjord-rnode","n_code_links":2,"syntology":{"n_ran":11,"n_unverified":4,"n_samples":15,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"BOSS","metrics":{"clean-FID":"20.13"},"uses_additional_data":false,"paper_date":"2023-12-27","paper":"/paper/bellman-optimal-step-size-straightening-of","paper_url":"https://arxiv.org/abs/2312.16414v3","paper_title":"Bellman Optimal Stepsize Straightening of Flow-Matching Models","code":"https://github.com/nguyenngocbaocmt02/boss","n_code_links":1,"syntology":{"n_ran":8,"n_unverified":6,"n_samples":14,"n_pointer_only_licence":14}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,885 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9623,"papers_checked":6885,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":2737},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-25T09:33:49+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":14,"rows_with_any_sample_ran":14,"distinct_papers_with_graph_line":14,"distinct_papers_with_any_sample_ran":14,"samples_over_distinct_papers":{"n_ran":181,"n_unverified":105,"n_samples":286,"n_pointer_only_licence":85,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":181,"n_unverified":105,"n_samples":286,"n_pointer_only_licence":85,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}