Research map / Image generation / Text-to-image diffusion
CompSlider: Compositional Slider for Disentangled Multiple-Attribute Image Generation
Zixin Zhu, Kevin Duarte, Mamshad Nayeem Rizve, Chengyuan Xu, Ratheesh Kalarot, Junsong Yuan
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 disentanglement loss L_clss (ablation) | CompSlider diffusion-loss only · Human sliders | entanglement (%): 36.68 -> 19.87 | Improved |
| add disentanglement loss L_clss (ablation) | CompSlider diffusion-loss only · Human sliders | continuity (%): 68.96 -> 76.49 | Improved |
| add disentanglement loss L_clss (ablation) | CompSlider diffusion-loss only · Human sliders | consistency (%): 63.21 -> 49.29 | Hurt |
| add structure loss L_st (ablation) | CompSlider with L_clss · Human sliders | consistency (%): 49.29 -> 90.95 | Improved |
| add structure loss L_st (ablation) | CompSlider with L_clss · Human sliders | entanglement (%): 19.87 -> 14.04 | Improved |
| add structure loss L_st (ablation) | CompSlider with L_clss · Human sliders | scope (%): 63.27 -> 59.02 | Hurt |
| add structure loss L_st (ablation) | CompSlider with L_clss · Human sliders | continuity (%): 76.49 -> 81.07 | Improved |
| lower structure-loss threshold from 0.5 to 0.1 (ablation) | CompSlider · Human sliders | continuity (%): 64.68 -> 81.07 | Improved |
| lower structure-loss threshold from 0.5 to 0.1 (ablation) | CompSlider · Human sliders | scope (%): 25.50 -> 59.02 | Improved |
| lower structure-loss threshold from 0.5 to 0.1 (ablation) | CompSlider · Human sliders | consistency (%): 96.44 -> 90.95 | Hurt |
| compositional slider vs ConceptSlider (ablation) | CompSlider · Human sliders | entanglement (%): 27.22 -> 14.04 | Improved |
| compositional slider vs ConceptSlider (user study) (ablation) | CompSlider · Human sliders | user preference (%): 34.16 -> 54.66 | Improved |
Closest papers by meaning
- Continuous, Subject-Specific Attribute Control in T2I Models by Identifying Semantic Directions (CVPR 2025 · similarity 0.87)
- XVerse: Consistent Multi-Subject Control of Identity and Semantic Attributes via DiT Modulation (NeurIPS 2025 · similarity 0.86)
- Composing Parts for Expressive Object Generation (CVPR 2025 · similarity 0.86)
- DisEnvisioner: Disentangled and Enriched Visual Prompt for Customized Image Generation (ICLR 2025 · similarity 0.85)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.85)
- Generating compositional scenes via Text-to-image RGBA Instance Generation (NeurIPS 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.
