Imagen video

Imagen video is an Android client for accessing Google's neural network. The platform allows you to create videos based on free-form text descriptions.

Development basics

The project is developed by the company's specialists and allows you to generate a video sequence without loading the original visual materials. Only a text block is used. The AI model is based on the Video U Net architecture, and the publicly available LAION 400M database is also used. The general principle of operation is similar to image creation. At the initial stage, it creates an outline - a video with low resolution and 16 frames. Then the algorithm in several passes complicates the video - new elements are added, frequency and resolution are increased, all artifacts are removed, sharpness is corrected. The final result is in HD format and 24 frames per second. At the time of publication the platform is functioning in test mode. According to information from the developers, work is underway to finalize the current functionality. It is also planned to update the set of filters to protect against unethical or prohibited content.

Application

It is sufficient to provide a text description as initial data. The user can specify the storyline, the presence of objects, the general style and so on. Since the service is at the testing stage, the length of clips is limited. This restriction is planned to be removed in the future. Recommendations on composing text queries can be found in the help section. You can also read the description of the neural network. If an error is detected, you can contact the developers, all shortcomings are promptly eliminated.

Features

• video generation by description;
• English-language interface;
• collection of working examples;
• platform at the testing stage;
• free download.