{"paper":{"title":"Aligning Text-to-Image Models using Human Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds.","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Craig Boutilier, Hao Liu, Kimin Lee, Mohammad Ghavamzadeh, Moonkyung Ryu, Olivia Watkins, Pieter Abbeel, Shixiang Shane Gu, Yuqing Du","submitted_at":"2023-02-23T17:34:53Z","abstract_excerpt":"Deep generative models have shown impressive results in text-to-image synthesis. However, current text-to-image models often generate images that are inadequately aligned with text prompts. We propose a fine-tuning method for aligning such models using human feedback, comprising three stages. First, we collect human feedback assessing model output alignment from a set of diverse text prompts. We then use the human-labeled image-text dataset to train a reward function that predicts human feedback. Lastly, the text-to-image model is fine-tuned by maximizing reward-weighted likelihood to improve "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our method generates objects with specified colors, counts and backgrounds more accurately than the pre-trained model.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Human feedback on image-text alignment is consistent enough to be captured by a learned reward function that generalizes to new prompts and does not introduce unintended biases during fine-tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A three-stage fine-tuning process uses human ratings to train a reward model and then improves text-to-image alignment by maximizing reward-weighted likelihood.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6cd0b1bf33b50df1bbf267ede71a4204cbb4f4fe5e7600cee531f564453ec6e1"},"source":{"id":"2302.12192","kind":"arxiv","version":1},"verdict":{"id":"88c2a63e-42c0-43e5-86ce-ba86a719a11a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:35:53.230683Z","strongest_claim":"Our method generates objects with specified colors, counts and backgrounds more accurately than the pre-trained model.","one_line_summary":"A three-stage fine-tuning process uses human ratings to train a reward model and then improves text-to-image alignment by maximizing reward-weighted likelihood.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Human feedback on image-text alignment is consistent enough to be captured by a learned reward function that generalizes to new prompts and does not introduce unintended biases during fine-tuning.","pith_extraction_headline":"Fine-tuning text-to-image models with human feedback improves accuracy on prompts specifying colors, counts, and backgrounds."},"references":{"count":26,"sample":[{"doi":"","year":null,"title":"A General Language Assistant as a Laboratory for Alignment","work_id":"a43f9ea0-01be-47d5-b8ee-a1a9f73381c5","ref_index":1,"cited_arxiv_id":"2112.00861","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:1607.07086 , year=","work_id":"a298def3-ff4b-4b72-9b55-5707acf335ba","ref_index":2,"cited_arxiv_id":"1607.07086","is_internal_anchor":true},{"doi":"","year":2005,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","ref_index":3,"cited_arxiv_id":"2204.05862","is_internal_anchor":true},{"doi":"","year":null,"title":"E., and Wang, W","work_id":"63b29782-c84e-4467-9583-aee36944de57","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion","work_id":"ca618c21-3ba6-448e-bd86-bcecff3cdeb5","ref_index":5,"cited_arxiv_id":"2208.01618","is_internal_anchor":true}],"resolved_work":26,"snapshot_sha256":"86521c9baf0ae9ec1490431a238d8f1e0c731923e8d77dff6e41c6f4033768ab","internal_anchors":17},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}