Research map / Image generation / Text-to-image diffusion
LCM-Lookahead for Encoder-based Text-to-Image Personalization
Rinon Gal, Or Lichter, Elad Richardson, Or Patashnik, Amit Bermano, Gal Chechik, Danny Cohen-Or
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 LoRA scaling to preserve LCM/non-LCM alignment during lookahead loss training (ablation) | LCM-Lookahead identity loss · FFHQ-5000 | ID similarity: 0.246 -> 0.345 | Improved |
| add SDS loss to preserve alignment (ablation) | LCM-Lookahead identity loss · FFHQ-5000 | ID similarity: 0.246 -> 0.267 | Improved |
| add consistency (LCM) loss to preserve alignment (ablation) | LCM-Lookahead identity loss · FFHQ-5000 | ID similarity: 0.246 -> 0.324 | Improved |
| use LoRA scaling over LCM loss (ablation) | alignment preservation · FFHQ-5000 | ID similarity: 0.324 -> 0.345 | Improved |
| no alignment preservation in lookahead loss (ablation) | LCM-Lookahead · FFHQ-5000 / Unsplash-50 | ID similarity: 0.345 -> 0.246 (FFHQ-5000) | Hurt |
| train on photorealistic-only data (CelebA or ConsiStory) instead of SDXL Turbo synthetic data (ablation) | encoder training · FFHQ-5000 | CLIP-T: 26.33 -> 25.81 (CelebA); 26.33 -> 24.87 (ConsiStory) | Hurt |
| use SDXL celebrity synthetic data instead of SDXL Turbo identities (ablation) | encoder training · FFHQ-5000 | ID similarity: 0.345 -> 0.334 | No clear effect |
Closest papers by meaning
- Harnessing the Power of MLLMs for Transferable Text-to-Image Person ReID (CVPR 2024 · similarity 0.83)
- InstantBooth: Personalized Text-to-Image Generation without Test-Time Finetuning (CVPR 2024 · similarity 0.82)
- AttnDreamBooth: Towards Text-Aligned Personalized Text-to-Image Generation (NeurIPS 2024 · similarity 0.81)
- FineStyle: Fine-grained Controllable Style Personalization for Text-to-image Models (NeurIPS 2024 · similarity 0.81)
- Semantic-Aware Data Augmentation for Text-to-Image Synthesis (AAAI 2024 · similarity 0.81)
- JeDi: Joint-Image Diffusion Models for Finetuning-Free Personalized Text-to-Image Generation (CVPR 2024 · similarity 0.80)
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.
