MT Forcing：基于自回归扩散的稳定一致多轮图像编辑
原标题：Paper page - MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion
AI 摘要
该论文提出 MT Forcing，一种面向基于自回归扩散的统一多模态模型（UMM）的多轮图像编辑训练范式，通过单轮分布匹配蒸馏（Single-Turn DMD）与教师强制引导（Teacher-Forcing Guidance）两种互补训练信号，缓解多轮编辑中的视觉质量退化与内容漂移问题。在 MTEdit-Bench 上，将 MT Forcing 应用于 SenseNova-U1.5 后，八轮平均指令遵循分数从 8.47 提升至 9.66，内容一致性从 7.09 提升至 8.40。模型权重已在 Hugging Face 发布。
正文节选
MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion Abstract MT Forcing improves multi-turn image editing by combining single-turn distribution matching distillation and teacher-forcing guidance to stabilize visual quality and preserve cross-turn consistency. Multi-turn image editing aims to modify an image according to a sequence of instructions, enabling more effective interactive visual content creation for users. However, existing methods often ex