VisionCreator-R1: A Reflection-Enhanced Native Visual-Generation Agentic Model
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915848733589504 |
|---|---|
| author | Lai, Jinxiang Zhao, Wenzhe Lu, Zexin Zhang, Hualei Yang, Qinyu Quan, Rongwei Li, Zhimin Shao, Shuai Guo, Song Lu, Qinglin |
| author_facet | Lai, Jinxiang Zhao, Wenzhe Lu, Zexin Zhang, Hualei Yang, Qinyu Quan, Rongwei Li, Zhimin Shao, Shuai Guo, Song Lu, Qinglin |
| contents | Visual content generation has advanced from single-image to multi-image workflows, yet existing agents remain largely plan-driven and lack systematic reflection mechanisms to correct mid-trajectory visual errors. To address this limitation, we propose VisionCreator-R1, a native visual generation agent with explicit reflection, together with a Reflection-Plan Co-Optimization (RPCO) training methodology. Through extensive experiments and trajectory-level analysis, we uncover reflection-plan optimization asymmetry in reinforcement learning (RL): planning can be reliably optimized via plan rewards, while reflection learning is hindered by noisy credit assignment. Guided by this insight, our RPCO first trains on the self-constructed VCR-SFT dataset with reflection-strong single-image trajectories and planning-strong multi-image trajectories, then co-optimization on VCR-RL dataset via RL. This yields our unified VisionCreator-R1 agent, which consistently outperforms Gemini2.5Pro on existing benchmarks and our VCR-bench covering single-image and multi-image tasks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_08812 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | VisionCreator-R1: A Reflection-Enhanced Native Visual-Generation Agentic Model Lai, Jinxiang Zhao, Wenzhe Lu, Zexin Zhang, Hualei Yang, Qinyu Quan, Rongwei Li, Zhimin Shao, Shuai Guo, Song Lu, Qinglin Computer Vision and Pattern Recognition Visual content generation has advanced from single-image to multi-image workflows, yet existing agents remain largely plan-driven and lack systematic reflection mechanisms to correct mid-trajectory visual errors. To address this limitation, we propose VisionCreator-R1, a native visual generation agent with explicit reflection, together with a Reflection-Plan Co-Optimization (RPCO) training methodology. Through extensive experiments and trajectory-level analysis, we uncover reflection-plan optimization asymmetry in reinforcement learning (RL): planning can be reliably optimized via plan rewards, while reflection learning is hindered by noisy credit assignment. Guided by this insight, our RPCO first trains on the self-constructed VCR-SFT dataset with reflection-strong single-image trajectories and planning-strong multi-image trajectories, then co-optimization on VCR-RL dataset via RL. This yields our unified VisionCreator-R1 agent, which consistently outperforms Gemini2.5Pro on existing benchmarks and our VCR-bench covering single-image and multi-image tasks. |
| title | VisionCreator-R1: A Reflection-Enhanced Native Visual-Generation Agentic Model |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.08812 |