A Video Quality User Study for Video Diffusion Models

Are you interested in the latest developments in video diffusion models? We invite you to participate in our user study, where you can watch a few video samples and provide your ratings. By participating, you will have the chance to experience the latest video diffusion models firsthand. The study should take as little as 1 minute to complete.

What Are We Evaluating?

This study aims to assess the quality of videos generated by two different diffusion-based generative models. We are focusing on key aspects such as spatial and temporal consistency, as well as multi-view coherence. While these models produce impressive results, they can also generate unexpected artifacts, such as strange morphing effects, abrupt changes in subjects, or unusual occlusions and geometries. We have created 50 videos for this study, each video is generated using identical text prompts across these two models, and we may expand the study to include more videos and additional models in the future.


🙂 Pretty Good	🤔 Subject changed and weird books.

Your Contribution Matters

The results of this study, along with the videos and prompts used, will be made publicly available as part of a dataset. This dataset will contribute to a research report aimed at advancing the understanding of video diffusion models. Your feedback is crucial in helping us identify and understand these phenomena. Rest assured, the study is fully anonymous, and no personal information will be collected.

How to Participate

You can join the study using any device, whether it is a phone, tablet, laptop, or desktop computer. For the best experience, we recommend using a larger screen to observe potential artifacts more clearly. Please note that some mobile devices may have energy-saving settings that prevent videos from autoplaying; if this happens, simply click on the video boxes to play them manually.

Your participation is invaluable to us. Thank you for contributing to the advancement of video diffusion technology. Feel free to contact me Zirui Wang (zirui dot wang @eng.ox.ac.uk) for additional feedbacks.