Research map / Image generation / Text-to-image diffusion
PositionIC: Unified Position and Identity Consistency for Image Customization
Junjie Hu, Tianyang Han, Kai Ma, Jialin Gao, Yang Song, Xianhua He, Junfeng Luo, Xiaoming Wei, Wenqiang 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 |
|---|---|---|---|
| add visibility-aware attention (VAA) restricting reference tokens to their bounding-box region (ablation) | PositionIC · DreamBench single-subject | DINO: 0.686 -> 0.823 | Improved |
| add visibility-aware attention (VAA) (ablation) | PositionIC · DreamBench single-subject | CLIP-I: 0.784 -> 0.846 | Improved |
| add visibility-aware attention (VAA) (ablation) | PositionIC · DreamBench single-subject | CLIP-T: 0.269 -> 0.269 | No clear effect |
| add VAA for position control (ablation) | PositionIC · qualitative position control | position control ability | Improved |
| add volumetric weight regulation (VWR) for overlapping objects (ablation) | PositionIC · overlap cases (qualitative) | occlusion handling | Improved |
| train on filtered PIC-98K instead of unfiltered PIC-400K or Subject200K (ablation) | PositionIC · DreamBench | subject fidelity | Improved |
| use PositionIC over UNO on single-subject DreamBench | PositionIC · DreamBench single-subject | DINO: 0.814 -> 0.823 | Improved |
| use PositionIC over UNO on multi-subject DreamBench | PositionIC · DreamBench multi-subject | DINO: 0.707 -> 0.771 | Improved |
| use PositionIC over Gligen on single-subject position control | PositionIC · PositionIC-Bench single-subject | AP70: IoU 0.808 -> 0.828; AP70 0.811 -> 0.761 | Mixed |
| use PositionIC over Gligen on multi-subject position control | PositionIC · PositionIC-Bench multi-subject | AP: 0.628 -> 0.701 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- Instance-Diffusion — InstanceDiffusion: Instance-level Control for Image Generation (CVPR 2024) · better on 2 of 2 shared results
- OmniGen — OmniGen: Unified Image Generation (CVPR 2025) · better on 0 of 1 shared result
- OminiControl — OminiControl: Minimal and Universal Control for Diffusion Transformer (ICCV 2025) · better on 1 of 1 shared result
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
- IC-Custom: Diverse Image Customization via In-Context Learning (ICLR 2026 · similarity 0.87)
- Customizable Image Synthesis with Multiple Subjects (NeurIPS 2023 · similarity 0.86)
- ContextGen: Contextual Layout Anchoring for Identity-Consistent Multi-Instance Generation (ICLR 2026 · similarity 0.86)
- Multi-party Collaborative Attention Control for Image Customization (CVPR 2025 · similarity 0.85)
- DreamRelation: Bridging Customization and Relation Generation (CVPR 2025 · similarity 0.84)
- SpotActor: Training-Free Layout-Controlled Consistent Image Generation (AAAI 2025 · 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.
