LUSD: Localized Update Score Distillation for Text-Guided Image Editing
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909671177060352 |
|---|---|
| author | Chinchuthakun, Worameth Saengja, Tossaporn Tritrong, Nontawat Rewatbowornwong, Pitchaporn Khungurn, Pramook Suwajanakorn, Supasorn |
| author_facet | Chinchuthakun, Worameth Saengja, Tossaporn Tritrong, Nontawat Rewatbowornwong, Pitchaporn Khungurn, Pramook Suwajanakorn, Supasorn |
| contents | While diffusion models show promising results in image editing given a target prompt, achieving both prompt fidelity and background preservation remains difficult. Recent works have introduced score distillation techniques that leverage the rich generative prior of text-to-image diffusion models to solve this task without additional fine-tuning. However, these methods often struggle with tasks such as object insertion. Our investigation of these failures reveals significant variations in gradient magnitude and spatial distribution, making hyperparameter tuning highly input-specific or unsuccessful. To address this, we propose two simple yet effective modifications: attention-based spatial regularization and gradient filtering-normalization, both aimed at reducing these variations during gradient updates. Experimental results show our method outperforms state-of-the-art score distillation techniques in prompt fidelity, improving successful edits while preserving the background. Users also preferred our method over state-of-the-art techniques across three metrics, and by 58-64% overall. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_11054 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LUSD: Localized Update Score Distillation for Text-Guided Image Editing Chinchuthakun, Worameth Saengja, Tossaporn Tritrong, Nontawat Rewatbowornwong, Pitchaporn Khungurn, Pramook Suwajanakorn, Supasorn Graphics Computer Vision and Pattern Recognition Machine Learning While diffusion models show promising results in image editing given a target prompt, achieving both prompt fidelity and background preservation remains difficult. Recent works have introduced score distillation techniques that leverage the rich generative prior of text-to-image diffusion models to solve this task without additional fine-tuning. However, these methods often struggle with tasks such as object insertion. Our investigation of these failures reveals significant variations in gradient magnitude and spatial distribution, making hyperparameter tuning highly input-specific or unsuccessful. To address this, we propose two simple yet effective modifications: attention-based spatial regularization and gradient filtering-normalization, both aimed at reducing these variations during gradient updates. Experimental results show our method outperforms state-of-the-art score distillation techniques in prompt fidelity, improving successful edits while preserving the background. Users also preferred our method over state-of-the-art techniques across three metrics, and by 58-64% overall. |
| title | LUSD: Localized Update Score Distillation for Text-Guided Image Editing |
| topic | Graphics Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2503.11054 |