Research map / Image generation / Editing and style transfer
The Consistency Critic: Correcting Inconsistencies in Generated Images via Reference-Guided Attentive Alignment
Ziheng Ouyang, Yiren Song, Yaoli Liu, Shihao Zhu, Qibin Hou, Ming-Ming Cheng, Mike Zheng Shou
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 |
|---|---|---|---|
| apply ImageCritic post-editing to XVerse outputs | XVerse · CriticBench | CLIP-I: 76.5 -> 79.9 | Improved |
| apply ImageCritic post-editing to MOSAIC outputs | MOSAIC · CriticBench | DreamSim: 35.2 -> 31.4 | Improved |
| apply ImageCritic to strong closed-source generators (Nano-Banana) | Nano-Banana · CriticBench | CLIP-I: 79.2 -> 79.8 | Improved |
| apply ImageCritic to Qwen-Image outputs | Qwen-Image · DreamBench++ | DINO: 61.8 -> 64.9 | Improved |
| apply ImageCritic to Dreamo outputs | Dreamo · DreamBench++ | CLIP-I: 83.3 -> 83.4 | No clear effect |
| add attention alignment loss to the fine-tuned model (ablation) | ImageCritic · CriticBench (avg over generators) | DINO: 68.1 -> 68.6 | Improved |
| add detail encoder to the fine-tuned model (ablation) | ImageCritic · CriticBench (avg over generators) | CLIP-I: 77.9 -> 78.3 | Improved |
| combine attention alignment loss and detail encoder (ablation) | ImageCritic · CriticBench (avg over generators) | DreamSim: 31.3 -> 29.8 | Improved |
| fine-tune Flux Kontext on the reference-degraded-target dataset (no extra modules) (ablation) | Flux Kontext · CriticBench (avg over generators) | CLIP-I: +0.3 CLIP-I, +0.4 DINO | Improved |
Closest papers by meaning
- Refine-by-Align: Reference-Guided Artifacts Refinement through Semantic Alignment (ICLR 2025 · similarity 0.88)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.86)
- GenScale: A Benchmark for Relative Object Scale in Image Generation and Editing (NeurIPS 2026 · similarity 0.86)
- Improving Editability in Image Generation with Layer-wise Memory (CVPR 2025 · similarity 0.86)
- ImagenHub: Standardizing the evaluation of conditional image generation models (ICLR 2024 · similarity 0.86)
- Factuality Matters: When Image Generation and Editing Meet Structured Visuals (ICLR 2026 · similarity 0.85)
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.
