Research map / Image generation / Text-to-image diffusion
DICE: Distilling Classifier-Free Guidance into Text Embeddings
Zhenyu Zhou, Defang Chen, Can Wang, Chun Chen, Siwei Lyu
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 |
|---|---|---|---|
| sharpen text embeddings with lightweight sharpener (CFG-distilled) | unguided SD15 sampling · MS-COCO | FID: 32.80 -> 22.22 | Improved |
| sharpen text embeddings | unguided SDXL · MS-COCO | FID: 61.19 -> 28.01 | Improved |
| sharpen text embeddings | unguided Pixart-α · MS-COCO | CLIP score: 25.30 -> 29.51 | Improved |
| sharpen text embeddings | unguided DreamShaper · MS-COCO | FID: 24.17 -> 30.36 | No clear effect |
| naive scaling of text embeddings instead of learned sharpener (ablation) | unguided sampling · MS-COCO | CLIP score: 21.99 -> 22.89 (SD15, vs DICE 28.54) | No clear effect |
| DICE versus GD on SD15 | DICE · MS-COCO | FID: 23.54 -> 22.22 | Improved |
| sharpen padding embedding versus semantic embedding (ablation) | DICE · qualitative | image quality: cosine similarity 0.23 (padding) vs 0.75 (semantic) | Improved |
| DICE versus guided sampling | DICE · MS-COCO (SDXL) | CLIP score: 32.10 -> 30.63 | Hurt |
Closest papers by meaning
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.89)
- Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance (CVPR 2024 · similarity 0.88)
- Diffusion Adaptive Text Embedding for Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Noise-free Score Distillation (ICLR 2024 · similarity 0.87)
- Get What You Want, Not What You Don't: Image Content Suppression for Text-to-Image Diffusion Models (ICLR 2024 · 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.
