{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:LHXJRE7HKN6X56T5NJROFKTU34","short_pith_number":"pith:LHXJRE7H","canonical_record":{"source":{"id":"2605.13852","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.GR","submitted_at":"2026-03-25T11:30:22Z","cross_cats_sorted":["cs.CV","cs.LG"],"title_canon_sha256":"95f3416c3193a9313f442088e23f7bf49e841137d5cbc45da33098c61f62f69e","abstract_canon_sha256":"0445d383dfea12380fa8f9a75b165a33c1e31b321c48076d0016219e271aa124"},"schema_version":"1.0"},"canonical_sha256":"59ee9893e7537d7efa7d6a62e2aa74df3aaa0d013e25b93c0c56bacc25d089ec","source":{"kind":"arxiv","id":"2605.13852","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13852","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13852v1","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13852","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"pith_short_12","alias_value":"LHXJRE7HKN6X","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"LHXJRE7HKN6X56T5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"LHXJRE7H","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:LHXJRE7HKN6X56T5NJROFKTU34","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13852","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.GR","submitted_at":"2026-03-25T11:30:22Z","cross_cats_sorted":["cs.CV","cs.LG"],"title_canon_sha256":"95f3416c3193a9313f442088e23f7bf49e841137d5cbc45da33098c61f62f69e","abstract_canon_sha256":"0445d383dfea12380fa8f9a75b165a33c1e31b321c48076d0016219e271aa124"},"schema_version":"1.0"},"canonical_sha256":"59ee9893e7537d7efa7d6a62e2aa74df3aaa0d013e25b93c0c56bacc25d089ec","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:19.594292Z","signature_b64":"EvYfNzN5K7rSqSdFbcikuc5hRipGMLlpQiX8SIK3BuAe28FMRQkg4k0AGwwnVZrKqmp3N9sGgpWTHpZNI+35Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"59ee9893e7537d7efa7d6a62e2aa74df3aaa0d013e25b93c0c56bacc25d089ec","last_reissued_at":"2026-05-17T23:39:19.593494Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:19.593494Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13852","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"BkPcsTkaUJiYFs5Z353akDJJ5pB19KVsPHWyjt9tyY3LW50NoecVtrVAyDK1dforJHLJSP2Vv7tEViRoVvTcAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T23:25:25.674827Z"},"content_sha256":"98136b61d7534b2dcb3aaf7e8ba1a85d1bd8163199b8d746e85f52c472b3af96","schema_version":"1.0","event_id":"sha256:98136b61d7534b2dcb3aaf7e8ba1a85d1bd8163199b8d746e85f52c472b3af96"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:LHXJRE7HKN6X56T5NJROFKTU34","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance.","cross_cats":["cs.CV","cs.LG"],"primary_cat":"cs.GR","authors_text":"Andrea Vedaldi, Egor Zakharov, Ido Sobol, Kihyuk Sohn, Max Bluvstein, Or Litany, Yoav Blum","submitted_at":"2026-03-25T11:30:22Z","abstract_excerpt":"We often aim to generate images that are both photorealistic and 3D-consistent, adhering to precise geometry, material, and viewpoint controls. Typically, this is achieved by fine-tuning an image generator, pre-trained on billions of real images, using renders of synthetic 3D assets, where annotations for control signals are available. While this approach can learn the desired controls, it often compromises the realism of the images due to domain gap between photographs and renders. We observe that this issue largely arises from the model learning an unintended association between the presence"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The domain gap largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images, which the co-variate and adapters can fully mitigate without losing control accuracy.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Realiz3D decouples visual domain from 3D controls in diffusion models via domain-aware residual adapters to enable photorealistic controllable generation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"95f473791748cf69019a56e51af3388ea869e5cb13162e38ba31aeb18153566e"},"source":{"id":"2605.13852","kind":"arxiv","version":1},"verdict":{"id":"b9cb9b5c-c482-4dfc-a05c-361a4b0ddd2a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T07:32:58.501336Z","strongest_claim":"We introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain.","one_line_summary":"Realiz3D decouples visual domain from 3D controls in diffusion models via domain-aware residual adapters to enable photorealistic controllable generation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The domain gap largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images, which the co-variate and adapters can fully mitigate without losing control accuracy.","pith_extraction_headline":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance."},"references":{"count":46,"sample":[{"doi":"","year":2021,"title":"Deep vit features as dense visual descriptors.arXiv preprint arXiv:2112.05814, 2(3):4","work_id":"32f5cddf-1446-4477-986d-22ac06fd151f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Meta 3D TextureGen: Fast and consistent texture generation for 3d objects","work_id":"e2a0a769-0f90-4eb3-abb4-0a512409f916","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Synthlight: Por- trait relighting with diffusion model by learning to re-render synthetic faces","work_id":"4af99926-44c6-4874-8a7f-a9b4412fa0f2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Still-moving: Customized video generation without customized video data.ACM Transactions on Graphics (TOG), 43(6):1–11, 2024","work_id":"3459ffeb-dff9-4ba1-8c52-4d021802027e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Ambient diffu- sion: Learning clean distributions from corrupted data","work_id":"23f34387-bce7-406e-97a6-82bd037e8699","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":46,"snapshot_sha256":"14dc3289f954e84468cf28b5ede727d618e2a7c8b7bb3d6e03aba50a7bf9d7a8","internal_anchors":5},"formal_canon":{"evidence_count":1,"snapshot_sha256":"a8a4b9d2626918f52826fe5fbd355e8220f18d6019aaa58052582dfc720ca299"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"b9cb9b5c-c482-4dfc-a05c-361a4b0ddd2a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Mr+FJZFdW1X11ejWIQo7MP9ZDrsx8bzOUcVxYaR/ahg+iMsGZfHeG91nvjESiov3hrlWpBsP7+hAlX34WzcvAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T23:25:25.676046Z"},"content_sha256":"0b63ef5dc2037d087e31fd7231d8a428456623f14f138c79f6be14d47a1779ed","schema_version":"1.0","event_id":"sha256:0b63ef5dc2037d087e31fd7231d8a428456623f14f138c79f6be14d47a1779ed"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/LHXJRE7HKN6X56T5NJROFKTU34/bundle.json","state_url":"https://pith.science/pith/LHXJRE7HKN6X56T5NJROFKTU34/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/LHXJRE7HKN6X56T5NJROFKTU34/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-26T23:25:25Z","links":{"resolver":"https://pith.science/pith/LHXJRE7HKN6X56T5NJROFKTU34","bundle":"https://pith.science/pith/LHXJRE7HKN6X56T5NJROFKTU34/bundle.json","state":"https://pith.science/pith/LHXJRE7HKN6X56T5NJROFKTU34/state.json","well_known_bundle":"https://pith.science/.well-known/pith/LHXJRE7HKN6X56T5NJROFKTU34/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:LHXJRE7HKN6X56T5NJROFKTU34","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"0445d383dfea12380fa8f9a75b165a33c1e31b321c48076d0016219e271aa124","cross_cats_sorted":["cs.CV","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.GR","submitted_at":"2026-03-25T11:30:22Z","title_canon_sha256":"95f3416c3193a9313f442088e23f7bf49e841137d5cbc45da33098c61f62f69e"},"schema_version":"1.0","source":{"id":"2605.13852","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13852","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13852v1","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13852","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"pith_short_12","alias_value":"LHXJRE7HKN6X","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"LHXJRE7HKN6X56T5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"LHXJRE7H","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:0b63ef5dc2037d087e31fd7231d8a428456623f14f138c79f6be14d47a1779ed","target":"graph","created_at":"2026-05-17T23:39:19Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The domain gap largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images, which the co-variate and adapters can fully mitigate without losing control accuracy."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Realiz3D decouples visual domain from 3D controls in diffusion models via domain-aware residual adapters to enable photorealistic controllable generation."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance."}],"snapshot_sha256":"95f473791748cf69019a56e51af3388ea869e5cb13162e38ba31aeb18153566e"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"a8a4b9d2626918f52826fe5fbd355e8220f18d6019aaa58052582dfc720ca299"},"paper":{"abstract_excerpt":"We often aim to generate images that are both photorealistic and 3D-consistent, adhering to precise geometry, material, and viewpoint controls. Typically, this is achieved by fine-tuning an image generator, pre-trained on billions of real images, using renders of synthetic 3D assets, where annotations for control signals are available. While this approach can learn the desired controls, it often compromises the realism of the images due to domain gap between photographs and renders. We observe that this issue largely arises from the model learning an unintended association between the presence","authors_text":"Andrea Vedaldi, Egor Zakharov, Ido Sobol, Kihyuk Sohn, Max Bluvstein, Or Litany, Yoav Blum","cross_cats":["cs.CV","cs.LG"],"headline":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.GR","submitted_at":"2026-03-25T11:30:22Z","title":"Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning"},"references":{"count":46,"internal_anchors":5,"resolved_work":46,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Deep vit features as dense visual descriptors.arXiv preprint arXiv:2112.05814, 2(3):4","work_id":"32f5cddf-1446-4477-986d-22ac06fd151f","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Meta 3D TextureGen: Fast and consistent texture generation for 3d objects","work_id":"e2a0a769-0f90-4eb3-abb4-0a512409f916","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Synthlight: Por- trait relighting with diffusion model by learning to re-render synthetic faces","work_id":"4af99926-44c6-4874-8a7f-a9b4412fa0f2","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Still-moving: Customized video generation without customized video data.ACM Transactions on Graphics (TOG), 43(6):1–11, 2024","work_id":"3459ffeb-dff9-4ba1-8c52-4d021802027e","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Ambient diffu- sion: Learning clean distributions from corrupted data","work_id":"23f34387-bce7-406e-97a6-82bd037e8699","year":2023}],"snapshot_sha256":"14dc3289f954e84468cf28b5ede727d618e2a7c8b7bb3d6e03aba50a7bf9d7a8"},"source":{"id":"2605.13852","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T07:32:58.501336Z","id":"b9cb9b5c-c482-4dfc-a05c-361a4b0ddd2a","model_set":{"reader":"grok-4.3"},"one_line_summary":"Realiz3D decouples visual domain from 3D controls in diffusion models via domain-aware residual adapters to enable photorealistic controllable generation.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Realiz3D decouples visual domain from control signals via a co-variate and residual adapters so diffusion models can apply 3D controls without adopting synthetic appearance.","strongest_claim":"We introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain.","weakest_assumption":"The domain gap largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images, which the co-variate and adapters can fully mitigate without losing control accuracy."}},"verdict_id":"b9cb9b5c-c482-4dfc-a05c-361a4b0ddd2a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:98136b61d7534b2dcb3aaf7e8ba1a85d1bd8163199b8d746e85f52c472b3af96","target":"record","created_at":"2026-05-17T23:39:19Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"0445d383dfea12380fa8f9a75b165a33c1e31b321c48076d0016219e271aa124","cross_cats_sorted":["cs.CV","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.GR","submitted_at":"2026-03-25T11:30:22Z","title_canon_sha256":"95f3416c3193a9313f442088e23f7bf49e841137d5cbc45da33098c61f62f69e"},"schema_version":"1.0","source":{"id":"2605.13852","kind":"arxiv","version":1}},"canonical_sha256":"59ee9893e7537d7efa7d6a62e2aa74df3aaa0d013e25b93c0c56bacc25d089ec","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"59ee9893e7537d7efa7d6a62e2aa74df3aaa0d013e25b93c0c56bacc25d089ec","first_computed_at":"2026-05-17T23:39:19.593494Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:19.593494Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"EvYfNzN5K7rSqSdFbcikuc5hRipGMLlpQiX8SIK3BuAe28FMRQkg4k0AGwwnVZrKqmp3N9sGgpWTHpZNI+35Cw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:19.594292Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13852","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:98136b61d7534b2dcb3aaf7e8ba1a85d1bd8163199b8d746e85f52c472b3af96","sha256:0b63ef5dc2037d087e31fd7231d8a428456623f14f138c79f6be14d47a1779ed"],"state_sha256":"978e7b9ffd1de497d324cd05bea8f403a3976bed7911f34c232c6dd4cd7ee38f"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4TLtYU3r9tOxr0XSVCwRtkB6f6lfmhC2oXp0+M9nPjAR6kBOmRHAbl1RhAbGzuCvycOQem4bsam6WuPbAlOUDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-26T23:25:25.680538Z","bundle_sha256":"ee8b1b39ea4b0f2be18cd8a4515e5394275f06ad97069eef24e9b3376e815f00"}}