Research map / Image generation / Text-to-image diffusion
RADIANCE: Relative Adaptive Denoising with IP-Adapter for Novel Concept Enhancement
Zi-Xiang Ni, Bo-Lun Huang, Teng-Fang Hsiao, Bo-Kai Ruan, Hong-Han Shuai
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 Bidirectional Scale Controller to fixed-scale guidance (ablation) | RADIANCE · RareBench (single-object) | overall score: 73.0 -> 89.9 | Improved |
| add Bidirectional Scale Controller (ablation) | RADIANCE · RareBench (single-object) | overall score: 86.8 -> 89.9 | Improved |
| use feedback-driven scaling instead of negative prompt (ablation) | RADIANCE · RareBench (single-object) | overall score: 83.4 -> 89.9 | Improved |
| start feedback at T_early=17 instead of 15 (ablation) | RADIANCE · RareBench (single-object) | overall score: 89.9 -> 88.1 | Hurt |
| start feedback at T_early=13 instead of 15 (ablation) | RADIANCE · RareBench (single-object) | overall score: 89.9 -> 89.1 | No clear effect |
| add Delayed Adapter Activation (ablation) | RADIANCE · RareBench (multi-object) | overall score: 62.7 -> 76.0 | Improved |
| add Layer-wise Alternating Guidance (ablation) | RADIANCE · RareBench (multi-object) | overall score: 71.0 -> 76.0 | Improved |
| apply RADIANCE to SD 3.5 | SD 3.5 · RareBench | overall score: 75.9 -> 84.7 | Improved |
| apply RADIANCE to SD 1.5 | SD 1.5 · RareBench | overall score: 33.9 -> 44.0 | Improved |
| RADIANCE versus prompt-switching R2F on SD 3.5 | RADIANCE · RareBench | overall score: 81.7 -> 84.7 | Improved |
| RADIANCE versus R2F on Multi-Relation | RADIANCE · RareBench | Multi-Relation score: 64.4 -> 63.1 | No clear effect |
| apply RADIANCE to SD 3.5 | SD 3.5 · T2I-CompBench | overall score: 74.2 -> 75.9 | 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.
- InitNO — InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (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
- Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified Text-to-Image Generation (NeurIPS 2026 · similarity 0.86)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.86)
- Concept Reachability in Diffusion Models: Beyond Dataset Constraints (ICML 2025 · similarity 0.86)
- CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models (NeurIPS 2025 · similarity 0.86)
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.86)
- HyperFields: Towards Zero-Shot Generation of NeRFs from Text (ICML 2024 · similarity 0.85)
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.
