**Image Conductor: Precision Control for Interactive Video Synthesis** **Authors:** Yaowei Li, Xintao Wang, Zhaoyang Zhang, Zhouxia Wang, Ziyang Yuan, Liangbin Xie, Yuexian Zou, Ying Shan **Project Page:** <https://liyaowei-stu.github.io/project/ImageConductor/> **Abstract:** Filmmaking and animation production often require sophisticated techniques for coordinating camera transitions and object movements, typically involving labor-intensive real-world capturing. Despite advancements in generative AI for video creation, achieving precise control over motion for interactive video asset generation remains challenging. To address this, we propose Image Conductor, a method for precise control of camera transitions and object movements to generate video assets from a single image. Our method involves a well-cultivated training strategy that separates distinct camera and object motions using camera LoRA weights and object LoRA weights. Additionally, we introduce a camera-free guidance technique during inference to enhance object movements while eliminating camera transitions. We also develop a trajectory-oriented video motion data curation pipeline for training. Quantitative and qualitative experiments demonstrate the effectiveness of our method in generating motion-controllable videos from images, advancing the practical application of interactive video synthesis. **Introduction:** Filmmaking and animation production often require precise coordination of camera transitions and object movements. Current workflows heavily rely on real-world capturing and 3D scan modeling, which are labor-intensive and costly. Recent work has explored AIGC-based filmmaking pipelines using diffusion models to generate video clip assets. However, generating dynamic video assets with precise control remains challenging. Image Conductor aims to address this by providing fine-grained control over camera transitions and object movements. **Approach:** 1. **Trajectory-Oriented Video Motion Data Construction:** We construct a high-quality video motion dataset with precise trajectory annotations to address the lack of such data. 2. **Motion-Aware Image-to-Video Architecture:** We use Animatediff and SparseCtrl as the foundational model for image-to-video generation. 3. **Controllable Motion Separation:** We introduce camera LoRA and object LoRA to separate and control camera transitions and object movements. 4. **Camera-Free Guidance:** We propose a camera-free guidance technique to enhance object movements while eliminating camera transitions. **Experiments:** 1. **Comparisons with State-of-the-Art Methods:** Image Conductor outperforms existing methods in both qualitative and quantitative evaluations. 2. **Personalized and Controllable Video Synthesis:** Our method can seamlessly integrate with open-source customization communities. 3. **Ablation Studies:** We validate the effectiveness of distinct LoRA weights and camera-free guidance. **Conclusion:** Image Conductor is a novel approach for precise and fine-grained control of camera transitions and object movements in interactive video synthesis. Our method advances the practical application of video-centric creative expression by providing robust and user-friendly motion control.**Image Conductor: Precision Control for Interactive Video Synthesis** **Authors:** Yaowei Li, Xintao Wang, Zhaoyang Zhang, Zhouxia Wang, Ziyang Yuan, Liangbin Xie, Yuexian Zou, Ying Shan **Project Page:** <https://liyaowei-stu.github.io/project/ImageConductor/> **Abstract:** Filmmaking and animation production often require sophisticated techniques for coordinating camera transitions and object movements, typically involving labor-intensive real-world capturing. Despite advancements in generative AI for video creation, achieving precise control over motion for interactive video asset generation remains challenging. To address this, we propose Image Conductor, a method for precise control of camera transitions and object movements to generate video assets from a single image. Our method involves a well-cultivated training strategy that separates distinct camera and object motions using camera LoRA weights and object LoRA weights. Additionally, we introduce a camera-free guidance technique during inference to enhance object movements while eliminating camera transitions. We also develop a trajectory-oriented video motion data curation pipeline for training. Quantitative and qualitative experiments demonstrate the effectiveness of our method in generating motion-controllable videos from images, advancing the practical application of interactive video synthesis. **Introduction:** Filmmaking and animation production often require precise coordination of camera transitions and object movements. Current workflows heavily rely on real-world capturing and 3D scan modeling, which are labor-intensive and costly. Recent work has explored AIGC-based filmmaking pipelines using diffusion models to generate video clip assets. However, generating dynamic video assets with precise control remains challenging. Image Conductor aims to address this by providing fine-grained control over camera transitions and object movements. **Approach:** 1. **Trajectory-Oriented Video Motion Data Construction:** We construct a high-quality video motion dataset with precise trajectory annotations to address the lack of such data. 2. **Motion-Aware Image-to-Video Architecture:** We use Animatediff and SparseCtrl as the foundational model for image-to-video generation. 3. **Controllable Motion Separation:** We introduce camera LoRA and object LoRA to separate and control camera transitions and object movements. 4. **Camera-Free Guidance:** We propose a camera-free guidance technique to enhance object movements while eliminating camera transitions. **Experiments:** 1. **Comparisons with State-of-the-Art Methods:** Image Conductor outperforms existing methods in both qualitative and quantitative evaluations. 2. **Personalized and Controllable Video Synthesis:** Our method can seamlessly integrate with open-source customization communities. 3. **Ablation Studies:** We validate the effectiveness of distinct LoRA weights and camera-free guidance. **Conclusion:** Image Conductor is a novel approach for precise and fine-grained control of camera transitions and object movements in interactive video synthesis. Our method advances the practical application of video-centric creative expression by providing robust and user-friendly motion control.