Research map / Image generation / Text-to-image diffusion
NULL-Text Inversion for Editing Real Images Using Guided Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add DDIM pivotal initialization before null-text optimization (ablation) | Null-text inversion · COCO validation image-caption pairs | reconstruction quality (PSNR) | Helped |
| Use one optimized null-text embedding per timestamp instead of a single global embedding (ablation) | Null-text inversion · COCO validation image-caption pairs | reconstruction quality | Helped |
| Add pivotal inversion when optimizing null-text embeddings (ablation) | Null-text optimization · COCO validation image-caption pairs | reconstruction quality | Helped |
| Optimize the conditional text embedding instead of the unconditional null-text embedding, while retaining a pivot (ablation) | Pivotal inversion · Prompt-to-Prompt editing | editing fidelity | Hurt |
| Optimize the conditional text embedding using random noise samples instead of one pivot trajectory (ablation) | Textual inversion · COCO validation image-caption pairs | reconstruction quality | Hurt |
| Use a random latent trajectory instead of DDIM inversion as the pivot (ablation) | Null-text inversion · COCO validation image-caption pairs | convergence speed | Hurt |
| Use one global null-text embedding across all timestamps (ablation) | Null-text inversion · COCO validation image-caption pairs | convergence speed: 7500 optimization steps; about 30 minutes | Hurt |
| Use a random, unaligned input caption rather than an image-aligned caption (ablation) | Null-text inversion · COCO validation image-caption pairs | reconstruction quality | No clear effect |
| Increase DDIM inversion guidance scale above 1 (ablation) | DDIM pivotal initialization · COCO validation image-caption pairs | editability | Hurt |
| Apply null-text inversion before SDEdit | SDEdit · Real-image text editing | fidelity to input image (LPIPS) | Helped |
| Use null-text inversion rather than Imagic | Real-image text editing · Stable Diffusion comparison to Imagic | fidelity to input image (LPIPS) | Helped |
Closest papers by meaning
- Inversion-Free Image Editing with Language-Guided Diffusion Models (CVPR 2024 · similarity 0.93)
- Lightning-Fast Image Inversion and Editing for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.92)
- Noise Map Guidance: Inversion with Spatial Context for Real Image Editing (ICLR 2024 · similarity 0.92)
- PnP Inversion: Boosting Diffusion-based Editing with 3 Lines of Code (ICLR 2024 · similarity 0.91)
- DCI: Dual-Conditional Inversion for Boosting Diffusion-Based Image Editing (NeurIPS 2025 · similarity 0.91)
- EditInfinity: Image Editing with Binary-Quantized Generative Models (NeurIPS 2025 · similarity 0.91)
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.