MT-EditFlow: Reinforcement Learning for Multi-Turn Image Editing with Flow Matching
Researchers have developed MT-EditFlow, a new framework that uses reinforcement learning and flow matching to improve multi-turn image editing. This approach addresses issues like error propagation and the all-or-nothing nature of single-turn edits. Experiments show MT-EditFlow significantly enhances performance across various models, notably boosting FLUX.1-Kontext-dev and outperforming models like Qwen-Image-Edit. AI
IMPACT Improves reliability and naturalness of human-AI collaboration in visual content creation.