Research map / Image generation / Text-to-image diffusion
Restoring Initial Noise Sensitivity in Text-to-Image Distillation through Geometric Alignment
Huayang Huang, Ruoyu Wang, Jinhui Zhao, Wei Deng, Daiguo Zhou, Jian Luan, Yu Wu, Ye Zhu
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 |
|---|---|---|---|
| add GAD (JVP matching regularizer) to LADD | LADD, SD2.1 · Seed identifiability | self-identifiability accuracy (%): 87.60% -> 92.40% | Improved |
| add GAD to LADD | LADD, SD2.1 · Layout control (800 COCO prompts) | AP: 5.0 -> 5.8 | Improved |
| add GAD to SiD | SiD, SANA · MS-COCO 1000 prompts | CLIP score: 32.75 -> 34.40 | Improved |
| add GAD to TDM | TDM, PixArt-alpha · MS-COCO 1000 prompts | PickScore: 22.1007 -> 22.2704 | Improved |
| add GAD to TDM | TDM, PixArt-alpha · zero-shot FID vs COCO-30k | FID: 25.3909 -> 26.3235 | Hurt |
| add GAD to LADD | LADD, SD2 · FID vs Teacher | FID: 11.2920 -> 7.8463 | Improved |
| add GAD to SiD | SiD, SANA · DrawBench zero-shot NoiseQuery transfer | HPSv2: 0.249 -> 0.289 | Improved |
| add GAD to TDM | TDM, PixArt-alpha · Trajectory deviation, 200 unseen prompts | final cumulative deviation: 0.491 -> 0.427 | Improved |
| replace Jacobian matching with noise augmentation (perturbed inputs, no matching target) (ablation) | TDM, PixArt-alpha · 1k COCO prompts | Vendi score: 2.7187 -> 2.6355 | Hurt |
| replace Jacobian matching with blind diversity repulsion loss (ablation) | TDM, PixArt-alpha · 1k COCO prompts | CLIP score: Vendi 3.0559 (highest), CLIP 33.42 -> 31.76 | Mixed |
| apply GAD only at mid-noise timesteps instead of full steps (ablation) | TDM, PixArt-alpha · GAD timesteps ablation | Vendi score: 2.5967 -> 2.7140 (full steps 2.7187) | Mixed |
| use increasing linear schedule for perturbation scale h instead of fixed h=1e-2 (ablation) | GAD on TDM, PixArt-alpha · scale schedule ablation | Vendi score: 2.7187 -> 2.7202 | No clear effect |
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.
- SwiftBrush — SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation (CVPR 2024) · better on 1 of 1 shared result
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
- Invertible Consistency Distillation for Text-Guided Image Editing in Around 7 Steps (NeurIPS 2024 · similarity 0.88)
- Consistent Flow Distillation for Text-to-3D Generation (ICLR 2025 · similarity 0.87)
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · similarity 0.87)
- Taming Mode Collapse in Score Distillation for Text-to-3D Generation (CVPR 2024 · similarity 0.87)
- SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation (CVPR 2024 · similarity 0.87)
- Diff-Instruct with Diffused Reward: Towards Principled One-step Generator RL (NeurIPS 2026 · similarity 0.87)
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.
