Research map / Image generation / Text-to-image diffusion
TARA: Token-Aware LoRA for Composable Personalization in Diffusion Models
Yuqi Peng, Lingtao Zheng, Yufeng Yang, Yi Huang, Mingfu Yan, Jianzhuang Liu, Shifeng Chen
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 token focus masking (TFM) (ablation) | DB-LoRA · DreamBooth multi-concept (average of 2-4 concepts) | DINO: 0.277 -> 0.374 | Improved |
| add token focus masking (TFM) (ablation) | DB-LoRA · DreamBooth multi-concept (average of 2-4 concepts) | CLIP-T: 0.295 -> 0.284 | Hurt |
| add token alignment loss (TAL) (ablation) | DB-LoRA · DreamBooth multi-concept (average of 2-4 concepts) | DINO: 0.277 -> 0.342 | Improved |
| add token alignment loss (TAL) (ablation) | DB-LoRA · DreamBooth multi-concept (average of 2-4 concepts) | CLIP-T: 0.295 -> 0.266 | Hurt |
| add TAL on top of TFM (ablation) | TARA · DreamBooth multi-concept (average of 2-4 concepts) | DINO: 0.374 -> 0.384 | Improved |
| add TAL on top of TFM (ablation) | TARA · DreamBooth multi-concept (average of 2-4 concepts) | CLIP-T: 0.284 -> 0.284 | No clear effect |
| TARA vs DB-LoRA for 4 concepts | TARA · DreamBooth multi-concept (4 concepts) | DINO: 0.253 -> 0.335 | Improved |
| TARA vs Mix-of-Show, no extra fusion training | TARA · DreamBooth multi-concept (average of 2-4 concepts) | DINO: 0.352 -> 0.384 | Mixed |
| use TARA instead of DB-LoRA on SDXL | TARA · DreamBooth single-concept after merging another LoRA | DINO: 0.356 -> 0.528 | Improved |
Closest papers by meaning
- LoRACLR: Contrastive Adaptation for Customization of Diffusion Models (CVPR 2025 · similarity 0.91)
- PaRa: Personalizing Text-to-Image Diffusion via Parameter Rank Reduction (ICLR 2025 · similarity 0.89)
- How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (NeurIPS 2024 · similarity 0.88)
- Orthogonal Adaptation for Modular Customization of Diffusion Models (CVPR 2024 · similarity 0.88)
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.88)
- Concept Weaver: Enabling Multi-Concept Fusion in Text-to-Image Models (CVPR 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.
