Research map / Vision-language and multimodal / Multimodal LLMs
MANGO: Unleashing Image Generation Capability of Unified Multimodal Models
Yi Wang, Mushui Liu, Wanggui He, Hanyang Yuan, Ziwei Huang, Guanghao Zhang, Wenkai Fang, Haoze Jiang, Shengxuming Zhang, Weilong Dai, Haofei Zhang, Mingli Song 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 Paint-CoT planning and acting (dense caption + layout) over best-of-two T2I (ablation) | MANGO 1.3B · GenEval | overall: 0.67 -> 0.73 | Improved |
| add reflection and correction on top of planning and acting (ablation) | MANGO 1.3B · GenEval | overall: 0.73 -> 0.77 | Improved |
| add reflection and correction on top of planning and acting (ablation) | MANGO 1.3B · T2I-CompBench | overall: 50.62 -> 54.85 | Improved |
| add Paint-CoT full process vs T2I twice (ablation) | MANGO 1.3B · GenEval | position: 0.40 -> 0.60 | Improved |
| use functionality-oriented transformers (FoT) instead of dense design (ablation) | MANGO · MS-COCO | FID: 11.3 -> 7.24 | Improved |
| use FoT instead of modality-oriented design (ablation) | MANGO · MS-COCO | FID: 9.56 -> 7.24 | Improved |
| use FoT instead of modality-oriented design (ablation) | MANGO · MS-COCO | CIDEr: 121.1 -> 126.5 | Improved |
| full Paint-CoT vs T2I generation on aesthetic score (ablation) | MANGO 1.3B · Aesthetic Score | aesthetic score: 5.98 -> 6.04 | No clear effect |
| full Paint-CoT vs T2I generation on human preference (ablation) | MANGO 1.3B · HPS v2 | win rate (%): 62.3 | Improved |
| run a second reflection-correction cycle (ablation) | MANGO 1.3B · GenEval | overall | No clear effect |
| scale backbone from Qwen2-0.5B to Qwen2-1.5B (MANGO-4B) | MANGO · GenEval | overall: 0.77 -> 0.82 | Improved |
| scale backbone from Qwen2-0.5B to Qwen2-1.5B (MANGO-4B) | MANGO · T2I-CompBench Complex | score: 42.78 -> 45.29 | 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.
- Show-o — Show-o: One Single Transformer to Unify Multimodal Understanding and Generation (ICLR 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
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.90)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.90)
- MMIG-Bench: Towards Comprehensive and Explainable Evaluation of Multi-Modal Image Generation Models (NeurIPS 2025 · similarity 0.90)
- Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasoners (ICML 2026 · similarity 0.89)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · similarity 0.88)
- MILR: Improving Multimodal Image Generation via Test-Time Latent Reasoning (ICLR 2026 · similarity 0.88)
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.
