IMAGAgent: Orchestrating Multi-Turn Image Editing via Constraint-Aware Planning and Reflection

arXiv:2603.29602 · cs.GR, cs.AI, cs.CV, cs.MA · Submitted 2026-02-12 · Read on arXiv

cs.GR, cs.AI, cs.CV, cs.MA

Submitted: 2026-02-12

Updated: 2026-10-11

Code: https://github.com/hackermmzz/IMAGAgent

Terminology

• Consistency — 1× in this paper · explained in 3 episodes
• Large Language Model (LLM) — 1× in this paper · explained in 3 episodes
• Semantic Drift — 1× in this paper · explained in 3 episodes
• Vision-Language Model (VLM) — 1× in this paper · explained in 3 episodes

Sources

• GPT-4 Technical Report
• Code2Video: A Code-centric Paradigm for Educational Video Generation
• DiffEdit: Diffusion-based semantic image editing with mask guidance
• GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning
• HQ-Edit: A High-Quality Dataset for Instruction-based Image Editing
• StyleTailor: Towards Personalized Fashion Styling via Hierarchical Negative Feedback
• ACE++: Instruction-Based Image Creation and Editing via Context-Aware Content Filling
• Visual Autoregressive Modeling for Instruction-Guided Image Editing
• SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations
• VINCIE: Unlocking In-context Image Editing from Video
• CLIP-Fields: Weakly Supervised Semantic Fields for Robotic Memory
• Visual Agents as Fast and Slow Thinkers
• OpenCUA: Open Foundations for Computer-Use Agents
• Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models
• MM-REACT: Prompting ChatGPT for Multimodal Reasoning and Action
• MLLM as Retriever: Interactively Learning Multimodal Retrieval for Embodied Agents
• In-Context Edit: Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer
• ChatVLA-2: Vision-Language-Action Model with Open-World Embodied Reasoning from Pretrained Knowledge
• Paper2Video: Automatic Video Generation from Scientific Papers

Related papers

• SCRIPT: Scalable Diffusion Policy with Multi-stage Training for Language-driven Physics-Based Humanoid Control
• CADReasoner: Iterative Program Editing for CAD Reverse Engineering
• QuadLink: Autoregressive Quad-Dominant Mesh Generation via Point-Relation Learning
• DrawVideo: Grounded and Faithful Multi-Shot Video Generation from Storyboard Keyframe Sketches
• MotionPersona: Real-Time Locomotion Control across Personas, Bodies, and Styles
• GaussianBench: Physics-Fidelity Evaluation for Gaussian Scene Representations