Research map / Image generation / Text-to-image diffusion
Discrete Noise Inversion for Next-scale Autoregressive Text-based Image Editing
Minh Quan Dao, Xiaoxiao He, Ligong Han, Ngan Nguyen, Amin Nobari, Han Zhang, Faez Ahmed, Viet Anh Nguyen, Dimitris N. Metaxas
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 |
|---|---|---|---|
| replace Regeneration with VARIN noise inversion | HART · PIE-Bench editing | Structure distance x10^3: 25.56 -> 11.46 | Improved |
| replace Regeneration with VARIN noise inversion | HART · PIE-Bench editing | PSNR: 20.45 -> 26.54 | Improved |
| replace Regeneration with VARIN noise inversion | HART · PIE-Bench editing | CLIP similarity (edited): 21.13 -> 21.49 | Hurt |
| use location-aware inversion with VARIN in HART instead of DICE on Paella | VARIN · PIE-Bench editing | CLIP similarity (whole): 23.79 -> 25.05 | Improved |
| VARIN with HART instead of DICE with Paella | VARIN · PIE-Bench editing | PSNR: 27.29 -> 26.54 | Hurt |
| VARIN instead of EditAR | VARIN · PIE-Bench editing | PSNR: 21.32 -> 26.54 | Improved |
| increase tau from 14 to 20 in LAI (ablation) | VARIN · PIE-Bench tau ablation | PSNR: 23.97 -> 27.30 | Improved |
| increase tau from 14 to 20 in LAI (ablation) | VARIN · PIE-Bench tau ablation | CLIP similarity (whole): 25.73 -> 24.79 | Hurt |
| increase Regeneration start scale s from 6 to 8 (ablation) | Regeneration · PIE-Bench Regeneration start scale | PSNR: 19.21 -> 20.45 | Improved |
| increase Regeneration start scale s from 6 to 8 (ablation) | Regeneration · PIE-Bench Regeneration start scale | CLIP similarity (whole): 26.83 -> 24.65 | Hurt |
| use VARIN instead of DDPM-Inversion (user study) | VARIN · User study | prompt agreement %: 52.35% -> 64.71% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EditAR — EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Edit the Bits, Diff the Codes: Bitwise Residual Editing for Visual Autoregressive Models (NeurIPS 2026 · similarity 0.91)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.90)
- Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations (ICLR 2025 · similarity 0.89)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.88)
- Fundamental Limits of Visual Autoregressive Transformers: Universal Approximation Abilities (ICML 2025 · similarity 0.88)
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 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.
