LMP: Leveraging Motion Prior in Zero-Shot Video Generation with Diffusion Transformer

In recent years, large-scale pre-trained diffusion transformer models have made significant progress in video generation. While current DiT models can produce high-definition, high-frame-rate, and highly diverse videos, there is a lack of fine-grained control over the video content. Controlling the motion of subjects in videos using only prompts is challenging, especially when it comes to describing complex movements. Further, existing methods fail to control the motion in image-to-video generation, as the subject in the reference image often differs from the subject in the reference video in terms of initial position, size, and shape. To address this, we propose the Leveraging Motion Prior (LMP) framework for zero-shot video generation. Our framework harnesses the powerful generative capabilities of pre-trained diffusion transformers to enable motion in the generated videos to reference user-provided motion videos in both text-to-video and image-to-video generation. To this end, we first introduce a foreground-background disentangle module to distinguish between moving subjects and backgrounds in the reference video, preventing interference in the target video generation. A reweighted motion transfer module is designed to allow the target video to reference the motion from the reference video. To avoid interference from the subject in the reference video, we propose an appearance separation module to suppress the appearance of the reference subject in the target video. We annotate the DAVIS dataset with detailed prompts for our experiments and design evaluation metrics to validate the effectiveness of our method. Extensive experiments demonstrate that our approach achieves state-of-the-art performance in generation quality, prompt-video consistency, and control capability. Our homepage is available at https://vpx-ecnu.github.io/LMP-Website/

FIND: Fine-tuning Initial Noise Distribution with Policy Optimization for Diffusion Models

In recent years, large-scale pre-trained diffusion models have demonstrated their outstanding capabilities in image and video generation tasks. However, existing models tend to produce visual objects commonly found in the training dataset, which diverges from user input prompts. The underlying reason behind the inaccurate generated results lies in the model's difficulty in sampling from specific intervals of the initial noise distribution corresponding to the prompt. Moreover, it is challenging to directly optimize the initial distribution, given that the diffusion process involves multiple denoising steps. In this paper, we introduce a Fine-tuning Initial Noise Distribution (FIND) framework with policy optimization, which unleashes the powerful potential of pre-trained diffusion networks by directly optimizing the initial distribution to align the generated contents with user-input prompts. To this end, we first reformulate the diffusion denoising procedure as a one-step Markov decision process and employ policy optimization to directly optimize the initial distribution. In addition, a dynamic reward calibration module is proposed to ensure training stability during optimization. Furthermore, we introduce a ratio clipping algorithm to utilize historical data for network training and prevent the optimized distribution from deviating too far from the original policy to restrain excessive optimization magnitudes. Extensive experiments demonstrate the effectiveness of our method in both text-to-image and text-to-video tasks, surpassing SOTA methods in achieving consistency between prompts and the generated content. Our method achieves 10 times faster than the SOTA approach. Our homepage is available at \url{https://github.com/vpx-ecnu/FIND-website}.

Motion-Zero: Zero-Shot Moving Object Control Framework for Diffusion-Based Video Generation

Recent large-scale pre-trained diffusion models have demonstrated a powerful generative ability to produce high-quality videos from detailed text descriptions. However, exerting control over the motion of objects in videos generated by any video diffusion model is a challenging problem. In this paper, we propose a novel zero-shot moving object trajectory control framework, Motion-Zero, to enable a bounding-box-trajectories-controlled text-to-video diffusion model. To this end, an initial noise prior module is designed to provide a position-based prior to improve the stability of the appearance of the moving object and the accuracy of position. In addition, based on the attention map of the U-net, spatial constraints are directly applied to the denoising process of diffusion models, which further ensures the positional and spatial consistency of moving objects during the inference. Furthermore, temporal consistency is guaranteed with a proposed shift temporal attention mechanism. Our method can be flexibly applied to various state-of-the-art video diffusion models without any training process. Extensive experiments demonstrate our proposed method can control the motion trajectories of objects and generate high-quality videos.