Research map / Image generation / Text-to-image diffusion
ContextGen: Contextual Layout Anchoring for Identity-Consistent Multi-Instance Generation
Ruihang Xu, Dewei Zhou, Fan Ma, Yi Yang
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 Contextual Layout Anchoring (composite layout image in context) (ablation) | ContextGen · LAMICBench++ | average score: 58.03 -> 62.15 (with ICA on all blocks); 64.66 with ICA on MID-19 | Improved |
| apply ICA only to the middle 19 DiT blocks instead of all blocks (ablation) | ContextGen · LAMICBench++ | average score: 62.15 -> 64.66 | Improved |
| apply ICA only to the last 19 blocks (ablation) | ContextGen · LAMICBench++ | average score: 64.66 -> 62.53 | Hurt |
| add ICA (reference tokens in context) on top of CLA-only (ablation) | ContextGen · LAMICBench++ overall vs overlapping layouts | average score: 64.22 (no ICA) vs 64.66 (ICA MID-19) | No clear effect |
| use instance-wise cumulative x and y position offsets (ablation) | ContextGen · LAMICBench++ | average score: 61.14 -> 64.66 | Improved |
| add an editing token (first position index set to 1) (ablation) | ContextGen · LAMICBench++ | average score: 64.25 -> 64.66 | No clear effect |
| add DPO fine-tuning with beta 1000 (ablation) | ContextGen (LoRA rank 256) · LAMICBench++ | average score: 62.55 -> 62.67 | Improved |
| add DPO fine-tuning (ablation) | ContextGen (LoRA rank 256) · LAMICBench++ | aesthetic quality (AES): 54.19 -> 57.10 (beta 1000) | Improved |
| add DPO fine-tuning (ablation) | ContextGen (LoRA rank 256) · LAMICBench++ | IDS: 32.37 -> 26.83 (beta 1000) | Hurt |
| port Grounding Resampler (MS-Diffusion style) to FLUX instead of CLA and ICA (ablation) | MS-FLUX · COCO-MIG | success rate (SR): 33.12 -> 0.38 | Hurt |
| use FLUX.1-Kontext rather than FLUX.1-Dev as backbone with attention masks (ablation) | ContextGen backbone · qualitative backbone selection | coherence and identity preservation | 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.
- MS-Diffusion — MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance (ICLR 2025) · 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
- InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention (NeurIPS 2025 · similarity 0.89)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.88)
- InstanceDiffusion: Instance-level Control for Image Generation (CVPR 2024 · similarity 0.87)
- OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (NeurIPS 2025 · similarity 0.87)
- SpotActor: Training-Free Layout-Controlled Consistent Image Generation (AAAI 2025 · similarity 0.87)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.87)
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.
