Research map / Vision-language and multimodal / Multimodal LLMs
GenAgent: Scaling Text-to-Image Generation via Agentic Multimodal Reasoning
Kaixun Jiang, Yuzheng Wang, Junjie Zhou, Pandeng Li, Zhihang Liu, Chen-Wei Xie, Zhaoyu Chen, Yun Zheng, 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 cold-start SFT on tool-invocation and reflection data (ablation) | GenAgent (Qwen2.5-VL-7B) · GenEval++ | overall score: 0.357 -> 0.507 | Improved |
| add cold-start SFT (ablation) | GenAgent (Qwen2.5-VL-7B) · WISE | overall score: 0.63 -> 0.64 | No clear effect |
| add cold-start SFT (ablation) | GenAgent (Qwen2.5-VL-7B) · Imagine | overall score: 6.258 -> 6.770 | Improved |
| add agentic RL after SFT (without pairwise reward) (ablation) | GenAgent · GenEval++ | overall score: 0.507 -> 0.543 | Improved |
| add pairwise reflection reward r_pair to RL (ablation) | GenAgent · GenEval++ | overall score: 0.543 -> 0.561 | Improved |
| add pairwise reflection reward r_pair to RL (ablation) | GenAgent · WISE | overall score: 0.68 -> 0.69 | Improved |
| add pairwise reflection reward r_pair to RL (ablation) | GenAgent · Imagine | overall score: 6.814 -> 6.825 | Improved |
| use tool invocation with FLUX.1-dev instead of single-pass rewrite only (ablation) | GenAgent · GenEval++ | overall score: 0.482 -> 0.561 | Improved |
| swap the FLUX.1-dev tool for stronger Qwen-Image at inference without retraining | GenAgent · GenEval++ | overall score: 0.561 -> 0.725 | Improved |
| swap the FLUX.1-dev tool for stronger Qwen-Image at inference without retraining | GenAgent · Imagine | overall score: 6.825 -> 7.794 | Improved |
| wrap FLUX.1-dev with GenAgent | FLUX.1-dev · WISE | overall score: 0.55 -> 0.69 | Improved |
| wrap FLUX.1-dev with GenAgent | FLUX.1-dev · GenEval++ | overall score: 0.325 -> 0.561 | Improved |
Closest papers by meaning
- Gen-Searcher: Reinforcing Agentic Search for Image Generation (NeurIPS 2026 · similarity 0.91)
- Generation Navigator: A State-Aware Agentic Framework for Image Generation (NeurIPS 2026 · similarity 0.90)
- GenArtist: Multimodal LLM as an Agent for Unified Image Generation and Editing (NeurIPS 2024 · similarity 0.89)
- Unify-Agent: A Unified Multimodal Agent for World-Grounded Image Synthesis (NeurIPS 2026 · similarity 0.89)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · similarity 0.88)
- Socratic-Geo: Synthetic Data Generation and Cross-Modal Geometric Reasoning via Multi-Agent Interaction (CVPR 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.
