{"ID":2829129,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2512.13672","arxiv_id":"2512.13672","title":"Directional Textual Inversion for Personalized Text-to-Image Generation","abstract":"Textual Inversion (TI) is an efficient approach to text-to-image personalization but often fails on complex prompts. We trace these failures to embedding norm inflation: learned tokens drift to out-of-distribution magnitudes, degrading prompt conditioning in pre-norm Transformers. Empirically, we show semantics are primarily encoded by direction in CLIP token space, while inflated norms harm contextualization; theoretically, we analyze how large magnitudes attenuate positional information and hinder residual updates in pre-norm blocks. We propose Directional Textual Inversion (DTI), which fixes the embedding magnitude to an in-distribution scale and optimizes only direction on the unit hypersphere via Riemannian SGD. We cast direction learning as MAP with a von Mises-Fisher prior, yielding a constant-direction prior gradient that is simple and efficient to incorporate. Across personalization tasks, DTI improves text fidelity over TI and TI-variants while maintaining subject similarity. Crucially, DTI's hyperspherical parameterization enables smooth, semantically coherent interpolation between learned concepts (slerp), a capability that is absent in standard TI. Our findings suggest that direction-only optimization is a robust and scalable path for prompt-faithful personalization. Code is available at https://github.com/kunheek/dti.","short_abstract":"Textual Inversion (TI) is an efficient approach to text-to-image personalization but often fails on complex prompts. We trace these failures to embedding norm inflation: learned tokens drift to out-of-distribution magnitudes, degrading prompt conditioning in pre-norm Transformers. Empirically, we show semantics are pri...","url_abs":"https://arxiv.org/abs/2512.13672","url_pdf":"https://arxiv.org/pdf/2512.13672v2","authors":"[\"Kunhee Kim\",\"NaHyeon Park\",\"Kibeom Hong\",\"Hyunjung Shim\"]","published":"2025-12-15T18:57:07Z","proceeding":"cs.LG","tasks":"[\"cs.LG\",\"cs.CV\"]","methods":"[\"Transformer\"]","has_code":false,"code_links":[{"ID":605933,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2829129,"paper_url":"https://arxiv.org/abs/2512.13672","paper_title":"Directional Textual Inversion for Personalized Text-to-Image Generation","repo_url":"https://github.com/kunheek/dti","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}