Research map / Image generation / Text-to-image diffusion
Decoding Correlation-Induced Misalignment in the Stable Diffusion Workflow for Text-to-Image Generation
Yunze Tong, Fengda Zhang, Didi Zhu, Jun Xiao, Kun Kuang
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 |
|---|---|---|---|
| fine-tune text-encoder self-attention with decorrelation loss | SD v1.5 · 30 correlated prompts | BERTScore: 75.74 -> 76.35 | Improved |
| fine-tune text-encoder self-attention with decorrelation loss | SD v1.5 · 30 correlated prompts | FID: 290.57 -> 243.11 | Improved |
| Ours vs PAG | SD v1.5 · 30 correlated prompts | BERTScore: 75.41 -> 76.35 | Improved |
| Ours vs AaE | SD v1.5 · 30 correlated prompts | FID: 276.27 -> 243.11 | Improved |
| Ours vs DisenDiff | SD v1.5 · 30 correlated prompts | FID: 308.16 -> 243.11 | Improved |
| Ours vs AaE on pair 3 | SD v1.5 · Object Pair 3 | FID: 155.90 -> 170.79 | Hurt |
| Ours vs pre-trained on pair 3 | SD v1.5 · Object Pair 3 | BERTScore: 75.14 -> 75.92 | No clear effect |
| Ours vs pre-trained in human study | SD v1.5 · User study | text-image alignment: 1.86 -> 3.60 | Improved |
| Ours vs pre-trained in human study | SD v1.5 · User study | image fidelity: 2.87 -> 3.79 | Improved |
| use fine-tuned model for conditional and pre-trained model for negative guidance instead of standard CFG (ablation) | Ours · qualitative prompts | object presence | Improved |
| use CFG with fine-tuned model only instead of dual-model guidance (ablation) | Ours · qualitative prompts | object presence and layout | Hurt |
| vary model-selection weight tau (1 to 10) (ablation) | Ours · trail and trash bin prompt | selected epoch: 12 -> 9 | No clear effect |
Closest papers by meaning
- Information Theoretic Text-to-Image Alignment (ICLR 2025 · similarity 0.90)
- Aligning Text to Image in Diffusion Models is Easier Than You Think (NeurIPS 2025 · similarity 0.90)
- Towards Transformer-Based Aligned Generation with Self-Coherence Guidance (CVPR 2025 · similarity 0.89)
- CoMat: Aligning Text-to-Image Diffusion Model with Image-to-Text Concept Matching (NeurIPS 2024 · similarity 0.89)
- Discriminative Probing and Tuning for Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- Text Embedding is Not All You Need: Attention Control for Text-to-Image Semantic Alignment with Text Self-Attention Maps (CVPR 2025 · similarity 0.89)
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.
