AutoStudio: Crafting Consistent Subjects in Multi-turn Interactive Image Generation
Junhao Cheng, Xi Lu, Hanhui Li, Khun Loun Zai, Baiqiao Yin, Yuhao Cheng, Yiqiang Yan, Xiaodan Liang*

[Project] [Code] [Paper]

• We propose a training-free multi-agent framework called AutoStudio. This framework stands out for its ability to maintain multi-subject consistency in on-the-fly multi-turn interactions with users, enabling it to accomplish various tasks such as open-ended story/manga book generation and multi-turn editing.

VisDiaHalBench: A Visual Dialogue Benchmark For Diagnosing Hallucination in Large Vision-Language Models
Qingxing Cao, Junhao Cheng, Xiaodan Liang*, Liang Lin*

[Project] [Code] [Paper]

• To investigate the hallucination problem of LVLMs when given long-term misleading textual history, we propose a novel visual dialogue hallucination evaluation benchmark VisDiaHalBench. The benchmark consists of samples with five-turn questions about an edited image and its original version. The benchmark is released in here.

TheaterGen: Character Management with LLM for Consistent Multi-turn Image Generation
Junhao Cheng, Baiqiao Yin, Kaixin Cai, Minbin Huang, Hanhui Li, Yuxin He, Xi Lu, Yue Li, Yifei Li, Yuhao Cheng, Yiqiang Yan, Xiaodan Liang*

[Project] [Code] [Paper]

• We propose TheaterGen, which is a training-free framework that utilizes a large language model to drive a text-to-image generation model, effectively addressing the issues of semantic consistency and contextual consistency in multi-turn image generation tasks without specialized training.

Integrating Domain Knowledge into Transformer for Short-Term Wind Power Forecasting
Junhao Cheng, Xing Luo*, Zhi Jin*

[Project] [Paper]

• We initially propose the DKFormer forecasting model, which integrates domain knowledge through three constraint modules that are crucial in data pre-processing, model training, and forecasting stages.