{"paper":{"title":"Adding Conditional Control to Text-to-Image Diffusion Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"ControlNet adds spatial controls like edges, depth, and human poses to pretrained text-to-image diffusion models.","cross_cats":["cs.AI","cs.GR","cs.HC","cs.MM"],"primary_cat":"cs.CV","authors_text":"Anyi Rao, Lvmin Zhang, Maneesh Agrawala","submitted_at":"2023-02-10T23:12:37Z","abstract_excerpt":"We present ControlNet, a neural network architecture to add spatial conditioning controls to large, pretrained text-to-image diffusion models. ControlNet locks the production-ready large diffusion models, and reuses their deep and robust encoding layers pretrained with billions of images as a strong backbone to learn a diverse set of conditional controls. The neural architecture is connected with \"zero convolutions\" (zero-initialized convolution layers) that progressively grow the parameters from zero and ensure that no harmful noise could affect the finetuning. We test various conditioning co"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We present ControlNet, a neural network architecture to add spatial conditioning controls to large, pretrained text-to-image diffusion models. ControlNet locks the production-ready large diffusion models, and reuses their deep and robust encoding layers pretrained with billions of images as a strong backbone to learn a diverse set of conditional controls.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The zero convolutions progressively grow parameters from zero and ensure that no harmful noise could affect the finetuning, allowing the pretrained backbone to remain intact while learning new controls.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ControlNet adds spatial conditioning controls to pretrained text-to-image diffusion models via zero convolutions for stable fine-tuning on small or large datasets.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"ControlNet adds spatial controls like edges, depth, and human poses to pretrained text-to-image diffusion models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ceb25131b2b913ce597fef9e72bf1b4d5aeb174d0fbc78ad74f35388a5233deb"},"source":{"id":"2302.05543","kind":"arxiv","version":3},"verdict":{"id":"c291990c-fe75-4b25-b563-e900c875fbcb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T22:38:53.240545Z","strongest_claim":"We present ControlNet, a neural network architecture to add spatial conditioning controls to large, pretrained text-to-image diffusion models. ControlNet locks the production-ready large diffusion models, and reuses their deep and robust encoding layers pretrained with billions of images as a strong backbone to learn a diverse set of conditional controls.","one_line_summary":"ControlNet adds spatial conditioning controls to pretrained text-to-image diffusion models via zero convolutions for stable fine-tuning on small or large datasets.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The zero convolutions progressively grow parameters from zero and ensure that no harmful noise could affect the finetuning, allowing the pretrained backbone to remain intact while learning new controls.","pith_extraction_headline":"ControlNet adds spatial controls like edges, depth, and human poses to pretrained text-to-image diffusion models."},"references":{"count":99,"sample":[{"doi":"","year":2020,"title":"Weight initialization in neural network, inspired by andrew ng, https://medium.com/@safrin1128/weight- initialization-in-neural-network-inspired-by-andrew-ng- e0066dc4a566, 2020","work_id":"3f9ca0ba-f454-40c9-b7c4-a550dd074663","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"In- trinsic dimensionality explains the effectiveness of language model fine-tuning","work_id":"59996242-57e2-4905-a5ee-ddff83d2fc2f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Only a matter of style: Age transformation using a style-based regression model","work_id":"f567145e-0ca7-4163-94d8-e24ef8c8b91a","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Hyperstyle: Stylegan inversion with hypernetworks for real image editing","work_id":"e058da51-899a-4555-b751-ec1e90b69a02","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Disco diffusion, https://github.com/alembics/disco- diffusion, 2022","work_id":"203b3fb8-4921-487f-9110-dda3201afe51","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":99,"snapshot_sha256":"0d6aa2f2d7e538e47c146e68c5e68fa5c1befbc4ff90b938d64e90da34937a3d","internal_anchors":6},"formal_canon":{"evidence_count":2,"snapshot_sha256":"46cd962887c3ef1518d72b679115b5e163ec5ad7c99142f28815d4356d358d27"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}