Research map / Image generation / Text-to-image diffusion
IC-Custom: Diverse Image Customization via In-Context Learning
Yaowei Li, Xiaoyu Li, Zhaoyang Zhang, Yuxuan Bian, Gan Liu, Xinyuan Li, Jiale Xu, Wenbo Hu, yating liu, Lingen Li, Jing Cai, Yuexian Zou and 2 more
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 input-layer training of DiT image and text layers (ablation) | IC-Custom · ProductBench (precise mask) | DINO-I: 62.00 -> 63.14 | Improved |
| add input-layer training (ablation) | IC-Custom · ProductBench (user-drawn mask) | DINO-I: 62.13 -> 63.28 | Improved |
| add real data to synthetic training data (ablation) | IC-Custom · ProductBench (precise mask) | DINO-I: 62.38 -> 63.14 | Improved |
| add user-drawn mask data in training (ablation) | IC-Custom · ProductBench (user-drawn mask) | DINO-I: 61.30 -> 63.28 | Improved |
| add task-oriented register tokens (ablation) | IC-Custom · ProductBench (precise mask) | DINO-I: 63.00 -> 63.14 | Improved |
| add boundary-aware positional embeddings (ablation) | IC-Custom · ProductBench (precise mask) | CLIP-I: 81.31 -> 81.92 | Improved |
| add task-oriented register tokens (ablation) | IC-Custom · ProductBench (user-drawn mask) | CLIP-T: 31.33 -> 31.80 | Improved |
| fine-tune on customization data versus zero-shot (ablation) | IC-Custom · ProductBench (precise mask) | DINO-I: 55.49 -> 63.14 | Improved |
| compare unified IC-Custom against ACE++ task-specific LoRA | IC-Custom · ProductBench (precise mask) | DINO-I: 60.68 -> 63.14 | Improved |
| compare IC-Custom against best baseline FLUX.1 workflow | IC-Custom · ProductBench (user-drawn mask) | DINO-I: 63.28 vs. 62.26 | Improved |
| compare IC-Custom against DreamO | IC-Custom · DreamBench (position-free) | DINO: 57.69 -> 65.67 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Insert Anything — Insert Anything: Image Insertion via In-Context Editing in DiT (AAAI 2026) · better on 2 of 2 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
- Multi-party Collaborative Attention Control for Image Customization (CVPR 2025 · similarity 0.87)
- DisEnvisioner: Disentangled and Enriched Visual Prompt for Customized Image Generation (ICLR 2025 · similarity 0.86)
- Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion models (NeurIPS 2024 · similarity 0.86)
- How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (NeurIPS 2024 · similarity 0.85)
- CustomContrast: A Multilevel Contrastive Perspective for Subject-Driven Text-to-Image Customization (AAAI 2025 · similarity 0.85)
- FreeCustom: Tuning-Free Customized Image Generation for Multi-Concept Composition (CVPR 2024 · similarity 0.84)
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.
