Research map / Image generation / Text-to-image diffusion
Editing Implicit Assumptions in Text-to-Image Diffusion Models
Hadas Orgad, Bahjat Kawar, Yonatan Belinkov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply TIME edit of cross-attention key/value projections | Stable Diffusion 1.4 · TIMED test | generality (%): 12.33% -> 69.04% | Improved |
| apply TIME edit | Stable Diffusion 1.4 · TIMED test | specificity (%): 90.13% -> 68.34% | Hurt |
| apply TIME edit | Stable Diffusion 1.4 · MS-COCO | FID: 12.67 -> 12.10 | No clear effect |
| optimize W_K in addition to W_V (ablation) | TIME · TIMED validation | harmonic mean (%): 64.24% -> 66.58% | Improved |
| add text prompt augmentations (ablation) | TIME · TIMED validation | harmonic mean (%): 66.26% -> 66.58% | Improved |
| increase lambda regularization from 0.1 to 100000 (ablation) | TIME · TIMED validation | generality (%): 67.80% -> 33.10% | Hurt |
| replace closed-form edit with text encoder finetuning (ablation) | TIME · TIMED validation | harmonic mean (%): 66.58% -> 52.13% | Hurt |
| apply TIME for gender debiasing with per-profession lambda | Stable Diffusion 1.4 · 35 professions | bias Delta: 0.57 -> 0.28 | Improved |
| debias all professions at once with a single lambda (multi-edit) (ablation) | TIME gender debiasing · 35 professions | bias Delta: 0.28 -> 0.48 | Hurt |
| edit many assumptions in bulk (ablation) | TIME multi-edit · 82 TIMED assumptions | specificity (%): 47% (efficacy 89%, generality 75%) | Hurt |
Closest papers by meaning
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.90)
- Exploring Low-Dimensional Subspace in Diffusion Models for Controllable Image Editing (NeurIPS 2024 · similarity 0.89)
- Adapting Diffusion Models for Improved Prompt Compliance and Controllable Image Synthesis (NeurIPS 2024 · similarity 0.89)
- Your Student is Better Than Expected: Adaptive Teacher-Student Collaboration for Text-Conditional Diffusion Models (CVPR 2024 · similarity 0.88)
- FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models (CVPR 2024 · similarity 0.88)
- Textualize Visual Prompt for Image Editing via Diffusion Bridge (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
