Research map / Image generation / Text-to-image diffusion
Steering Guidance for Personalized Text-to-Image Diffusion Models
Sunghyun Park, Seokeon Choi, Hyoungwoo Park, Sungrack Yun
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 |
|---|---|---|---|
| use pre-trained model with null prompt as the weak model instead of the fine-tuned unconditional model (personalization guidance vs CFG) | DB-LoRA on SANA · ViCo | DINO: 0.4819 -> 0.5366 | Improved |
| use personalization guidance instead of CFG | DB-LoRA on SD 2.1 · ViCo | CLIP-T: 0.3323 -> 0.3254 | Hurt |
| lower the weak-model weight interpolation omega from 1.0 toward 0.6 (ablation) | Personalization guidance on SD 1.5 DB-LoRA · ViCo | DINO: 0.4701 -> 0.5024 | Improved |
| increase omega toward 1.0 (more fine-tuned weak model) (ablation) | Personalization guidance on SD 1.5 DB-LoRA · ViCo | CLIP-T: 0.3260 -> 0.3345 | Improved |
| increase omega on SANA DB-LoRA (ablation) | Personalization guidance on SANA · ViCo | DINO: 0.5366 -> 0.4819 | Hurt |
| add AutoGuidance on top of CFG | DB-LoRA SD 1.5 · ViCo | DINO: 0.4701 -> 0.4831 | Improved |
| increase CFG guidance scale lambda vs personalization guidance (ablation) | SANA DB-LoRA · ViCo | DINO | Mixed |
| combine personalization guidance with CFG++ | DB-LoRA SD 2.1 · ViCo | DINO: 0.5105 -> 0.5385 | Improved |
| apply personalization guidance to Diffusion-DPO | Diffusion-DPO SD 1.5 · PartiPrompts | PickScore: 21.53 -> 21.58 | Improved |
| apply personalization guidance to PairCustom style LoRA | PairCustom · StyleAlign prompts | DreamSim: 0.7993 -> 0.7344 | Improved |
| personalization guidance versus CFG in DB-LoRA training steps (ablation) | SANA DB-LoRA · ViCo | subject and text fidelity | Improved |
| personalization guidance adds no overhead versus AG/SAG | SD 2.1 DB-LoRA+TI · Inference cost | inference time: 21.2 secs -> 14.5 secs | Improved |
Closest papers by meaning
- PaRa: Personalizing Text-to-Image Diffusion via Parameter Rank Reduction (ICLR 2025 · similarity 0.92)
- APT: Adaptive Personalized Training for Diffusion Models with Limited Data (CVPR 2025 · similarity 0.90)
- Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion models (NeurIPS 2024 · similarity 0.90)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.90)
- Navigating Text-To-Image Customization: From LyCORIS Fine-Tuning to Model Evaluation (ICLR 2024 · similarity 0.90)
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 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.
