Research map / Image generation / Text-to-image diffusion
RealCustom++: Representing Images as Real Textual Word for Real-Time Customization
Zhendong Mao, Mengqi Huang, Fei Ding, Mingcong Liu, Qian He, Yongdong Zhang
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 low-res cross-attn plus high-res self-attn mask guidance instead of conference version (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 32.20 -> 33.43 | Improved |
| use low-res cross-attn plus high-res self-attn mask guidance instead of conference version (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 84.23 -> 87.32 | Improved |
| use high-resolution instead of all-resolution self-attention for the mask (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 31.08 -> 33.43 | Improved |
| add maximum mask normalization (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 30.26 -> 33.43 | Improved |
| raise Top-K ratio from 0.1 to 0.2 (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 84.43 -> 87.32 | Improved |
| raise Top-K ratio from 0.2 to 0.5 (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 33.43 -> 31.08 | Hurt |
| add early stop mask regularization (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 86.67 -> 87.32 | Improved |
| add cross-scale mechanism to the projector (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 86.12 -> 87.32 | Improved |
| use cross-scale token concat + cross-layer element add instead of cross-scale element add (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 29.94 -> 33.43 | Improved |
| add curriculum data strategy instead of 50% random mixing (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 85.33 -> 86.11 | Improved |
| use curriculum cropping instead of fixed random cropping (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-I: 86.82 -> 87.32 | Improved |
| add multiview data to text-image data (curriculum) (ablation) | RealCustom++ · DreamBench SDXL | CLIP-B-T: 30.02 -> 32.10 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MS-Diffusion — MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance (ICLR 2025) · better on 8 of 8 shared results
- RealCustom — RealCustom: Narrowing Real Text Word for Real-Time Open-Domain Text-to-Image Customization (CVPR 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RealCustom: Narrowing Real Text Word for Real-Time Open-Domain Text-to-Image Customization (CVPR 2024 · similarity 0.95)
- CustomContrast: A Multilevel Contrastive Perspective for Subject-Driven Text-to-Image Customization (AAAI 2025 · similarity 0.88)
- Large-Scale Text-to-Image Model with Inpainting is a Zero-Shot Subject-Driven Image Generator (CVPR 2025 · similarity 0.88)
- Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion models (NeurIPS 2024 · similarity 0.88)
- AttnDreamBooth: Towards Text-Aligned Personalized Text-to-Image Generation (NeurIPS 2024 · similarity 0.88)
- MC^2: Multi-concept Guidance for Customized Multi-concept Generation (CVPR 2025 · similarity 0.86)
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.
