{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "peebles-xie-diffusion-transformer-backbone-2022",
    "title": "Peebles and Xie: diffusion transformers generate images from latent patches",
    "summary": "Peebles and Xie replace the U-Net denoiser in latent image diffusion with a conditioned transformer, reporting scalable class-conditional ImageNet generation through model size and latent-patch token count.\n\nDiffusion models generate images by repeatedly removing noise. Their usual denoising backbone was a convolutional U-Net. Peebles and Xie keep the established diffusion process but replace that backbone with a transformer that reads patches of compressed image representations. They compare ways to supply the noise level and image class, finding that adaptive layer normalization with zero-initialized residual gates works well in their experiments. Increasing model size or reducing patch size improves their chosen benchmark results within the studied configurations. Their best guided model reports an ImageNet256 FID of2.27. This expands the design options for diffusion generation; it does not invent diffusion, transformers or image autoencoders. Reported quality depends on training, decoder, guidance and evaluation settings, and the method still requires repeated denoising plus image decoding.",
    "boundary": "The work replaces the stated image-diffusion backbone with a transformer and evaluates scaling under specified training, data and guidance settings. Reported compute/quality correlations are not a universal scaling law or guarantee across all modalities. Data preparation, autoencoding and sampling costs remain. WRS has not independently replayed the experiments.",
    "area": "Generative image modeling architectures",
    "topic": "Mathematics and computing",
    "year": 2022,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "peebles-xie-diffusion-transformer-backbone-2022",
    "target": {
      "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
      "version": 1,
      "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
    },
    "humans": [
      "William Peebles",
      "Saining Xie"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed contributors to this scoped work. Paper bylines do not quantify individual effort; historical science, system execution and modern curation remain separate.",
    "agents": [],
    "related": [
      "dosovitskiy-et-al-vit-2020",
      "denoising-diffusion"
    ],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "wrs24-dit-iccv2023-main",
        "title": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper",
        "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf"
      },
      {
        "id": "wrs24-dit-iccv2023-supplement",
        "title": "DiT officialICCV2023supplement: implementation, complete model/decoder tables andsamplecaptions",
        "url": "https://openaccess.thecvf.com/content/ICCV2023/supplemental/Peebles_Scalable_Diffusion_Models_ICCV_2023_supplemental.pdf"
      },
      {
        "id": "wrs24-dit-arxiv-identity",
        "title": "OfficialarXiv DiT title/byline and2022first/2023revision history",
        "url": "https://arxiv.org/abs/2212.09748v2"
      }
    ],
    "points": "65000",
    "published_grade": {
      "id": "peebles-xie-diffusion-transformer-backbone-2022",
      "entry_id": "peebles-xie-diffusion-transformer-backbone-2022",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
        "version": 1,
        "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
      },
      "what_changed": "A conditioned latent-patch transformer backbone with adaLN-Zero and controlledsize/patch ablations broadens diffusion architectural options while retaining establisheddiffusion/VAE/ViT methods.",
      "scope": "A conditioned latent-patch transformer backbone with adaLN-Zero and controlledsize/patch ablations broadens diffusion architectural options while retaining establisheddiffusion/VAE/ViT methods. The work replaces the stated image-diffusion backbone with a transformer and evaluates scaling under specified training, data and guidance settings. Reported compute/quality correlations are not a universal scaling law or guarantee across all modalities. Data preparation, autoencoding and sampling costs remain. WRS has not independently replayed the experiments.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
        "version": 1,
        "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-03T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "William Peebles",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper",
            "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Saining Xie",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper",
            "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper",
            "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "DiT officialICCV2023supplement: implementation, complete model/decoder tables andsamplecaptions",
            "url": "https://openaccess.thecvf.com/content/ICCV2023/supplemental/Peebles_Scalable_Diffusion_Models_ICCV_2023_supplemental.pdf"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-2",
          "kind": "source_reference",
          "source": {
            "citation": "OfficialarXiv DiT title/byline and2022first/2023revision history",
            "url": "https://arxiv.org/abs/2212.09748v2"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "The work replaces the stated image-diffusion backbone with a transformer and evaluates scaling under specified training, data and guidance settings. Reported compute/quality correlations are not a universal scaling law or guarantee across all modalities. Data preparation, autoencoding and sampling costs remain. WRS has not independently replayed the experiments.",
        "domain": "Generative image modeling architectures"
      },
      "sources": [
        {
          "citation": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper",
          "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf"
        },
        {
          "citation": "DiT officialICCV2023supplement: implementation, complete model/decoder tables andsamplecaptions",
          "url": "https://openaccess.thecvf.com/content/ICCV2023/supplemental/Peebles_Scalable_Diffusion_Models_ICCV_2023_supplemental.pdf"
        },
        {
          "citation": "OfficialarXiv DiT title/byline and2022first/2023revision history",
          "url": "https://arxiv.org/abs/2212.09748v2"
        }
      ],
      "statement": "# Peebles and Xie: diffusion transformers generate images from latent patches\n\nPeebles and Xie replace the U-Net denoiser in latent image diffusion with a conditioned transformer, reporting scalable class-conditional ImageNet generation through model size and latent-patch token count.\n\nDiffusion models generate images by repeatedly removing noise. Their usual denoising backbone was a convolutional U-Net. Peebles and Xie keep the established diffusion process but replace that backbone with a transformer that reads patches of compressed image representations. They compare ways to supply the noise level and image class, finding that adaptive layer normalization with zero-initialized residual gates works well in their experiments. Increasing model size or reducing patch size improves their chosen benchmark results within the studied configurations. Their best guided model reports an ImageNet256 FID of2.27. This expands the design options for diffusion generation; it does not invent diffusion, transformers or image autoencoders. Reported quality depends on training, decoder, guidance and evaluation settings, and the method still requires repeated denoising plus image decoding.\n\n## Scientific scope\n\nThe work replaces the stated image-diffusion backbone with a transformer and evaluates scaling under specified training, data and guidance settings. Reported compute/quality correlations are not a universal scaling law or guarantee across all modalities. Data preparation, autoencoding and sampling costs remain. WRS has not independently replayed the experiments.\n\n## People and systems\n\nWilliam Peebles; Saining Xie\n\n\n\n## Original sources\n\n- [Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV2023 acceptedpaper](https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf)\n- [DiT officialICCV2023supplement: implementation, complete model/decoder tables andsamplecaptions](https://openaccess.thecvf.com/content/ICCV2023/supplemental/Peebles_Scalable_Diffusion_Models_ICCV_2023_supplemental.pdf)\n- [OfficialarXiv DiT title/byline and2022first/2023revision history](https://arxiv.org/abs/2212.09748v2)\n",
      "summary": "Peebles and Xie replace the U-Net denoiser in latent image diffusion with a conditioned transformer, reporting scalable class-conditional ImageNet generation through model size and latent-patch token count.\n\nDiffusion models generate images by repeatedly removing noise. Their usual denoising backbone was a convolutional U-Net. Peebles and Xie keep the established diffusion process but replace that backbone with a transformer that reads patches of compressed image representations. They compare ways to supply the noise level and image class, finding that adaptive layer normalization with zero-initialized residual gates works well in their experiments. Increasing model size or reducing patch size improves their chosen benchmark results within the studied configurations. Their best guided model reports an ImageNet256 FID of2.27. This expands the design options for diffusion generation; it does not invent diffusion, transformers or image autoencoders. Reported quality depends on training, decoder, guidance and evaluation settings, and the method still requires repeated denoising plus image decoding.",
      "title": "Peebles and Xie: diffusion transformers generate images from latent patches"
    },
    "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
      "version": 1,
      "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:8c8d727efe8f354c083925845730c8dad9fe6d146b4819c21263f2b8d2b2f902"
  },
  "target": {
    "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
    "version": 1,
    "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "peebles-xie-diffusion-transformer-backbone-2022",
    "entry_id": "peebles-xie-diffusion-transformer-backbone-2022",
    "revision": 1,
    "title": "Peebles and Xie: diffusion transformers generate images from latent patches",
    "context": "Generative image modeling architectures",
    "track": "Generative image modeling architectures",
    "band": "A",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-04",
    "evidence_cutoff": "2026-10-04",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "A conditioned latent-patch transformer backbone with adaLN-Zero and controlledsize/patch ablations broadens diffusion architectural options while retaining establisheddiffusion/VAE/ViT methods.",
    "scope": "A conditioned latent-patch transformer backbone with adaLN-Zero and controlledsize/patch ablations broadens diffusion architectural options while retaining establisheddiffusion/VAE/ViT methods. The work replaces the stated image-diffusion backbone with a transformer and evaluates scaling under specified training, data and guidance settings. Reported compute/quality correlations are not a universal scaling law or guarantee across all modalities. Data preparation, autoencoding and sampling costs remain. WRS has not independently replayed the experiments.",
    "evidence_summary": "Source-linked, scoped account of the original work. The published scope states the scientific assumptions and remaining verification limits. WRS has not independently certified the full proof, experimental result or system benchmark.",
    "references": [
      {
        "title": "Peebles and Xie, Scalable Diffusion Models with Transformers; CVF ICCV 2023 acceptedpaper",
        "url": "https://openaccess.thecvf.com/content/ICCV2023/papers/Peebles_Scalable_Diffusion_Models_with_Transformers_ICCV_2023_paper.pdf",
        "locator": "Original linked work. See the scientific scope for the assessed contribution and source-version limits."
      },
      {
        "title": "Di T official ICCV 2023 supplement: implementation, complete model/decoder tables andsamplecaptions",
        "url": "https://openaccess.thecvf.com/content/ICCV2023/supplemental/Peebles_Scalable_Diffusion_Models_ICCV_2023_supplemental.pdf",
        "locator": "Original linked work. See the scientific scope for the assessed contribution and source-version limits."
      },
      {
        "title": "Officialar Xiv Di T title/byline and 2022 first/2023 revision history",
        "url": "https://arxiv.org/abs/2212.09748v2",
        "locator": "Original linked work. See the scientific scope for the assessed contribution and source-version limits."
      }
    ],
    "roles": [
      {
        "who": "William Peebles",
        "did": "Original source-listed scientific author of this scoped work; byline does not determine effort fractions.",
        "credit": "SCIENCE"
      },
      {
        "who": "Saining Xie",
        "did": "Original source-listed scientific author of this scoped work; byline does not determine effort fractions.",
        "credit": "SCIENCE"
      }
    ],
    "target": {
      "record_id": "urn:uuid:a84f0d5b-90ef-48f9-a9aa-db41b130782f",
      "version": 1,
      "content_hash": "sha256:75e63fc7fadc3eacaff3afb2d671f58e5e2a2335f48cfd7e3b8e28cffa058c0d"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "65000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}