Research map / Image generation / Text-to-image diffusion
SCALE: Semantic-Calibrated Guidance Enhancement for Prompt-Faithful Diffusion
Tianhang Lu, Sudong Cai, Bingzhi Chen, Shao-Dong Shen, Chunting Liu, Longguang Wang, Bing Wang
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 SCALE guidance | SDXL · DrawBench | VQAScore: 0.7417 -> 0.7889 | Improved |
| add SCALE guidance | SDXL · GenEval | VQAScore: 0.7998 -> 0.8345 | Improved |
| add SCALE guidance | SDXL · T2I-CompBench | VQAScore: 0.7192 -> 0.7593 | Improved |
| add SCALE guidance | SDXL · GenEval overall accuracy | accuracy (%): 53.44% -> 56.41% | Improved |
| add SCALE guidance | SDXL · GenEval Two object | accuracy (%): 76.77% -> 71.72% | Hurt |
| scale both semantic and orthogonal components equally instead of only the semantic axis (ablation) | SCALE · DrawBench | ImageReward: 0.8305 -> -0.1666 | Hurt |
| preserve orthogonal component instead of discarding it (Parallel-Only) (ablation) | SCALE · DrawBench | VQAScore: 0.1446 -> 0.7889 | Improved |
| keep semantic component instead of discarding it (Perp-Only) (ablation) | SCALE · DrawBench | VQAScore: 0.2900 -> 0.7889 | Improved |
| apply SCALE on top of Diffusion-DPO | SDXL-DPO · DrawBench | VQAScore: 0.7685 -> 0.7955 | Improved |
| apply SCALE on top of Diffusion-DPO | SDXL-DPO · GenEval | ImageReward: 0.9856 -> 1.0278 | Improved |
| use SCALE instead of DAS reward guidance | DAS · DrawBench | VQAScore: 0.7483 -> 0.7889 | Improved |
| add SCALE to SDXL sampling (latency) | SDXL · inference time ratio | ratio: 1.0000 -> 1.0038 | No clear effect |
Closest papers by meaning
- Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation (ICML 2025 · similarity 0.91)
- Noise Diffusion for Enhancing Semantic Faithfulness in Text-to-Image Synthesis (CVPR 2025 · similarity 0.90)
- Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance (CVPR 2024 · similarity 0.90)
- Towards Transformer-Based Aligned Generation with Self-Coherence Guidance (CVPR 2025 · similarity 0.89)
- Improving Long-Text Alignment for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.89)
- Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion models (NeurIPS 2024 · 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.
