Brother
Professional
- Messages
- 2,590
- Reaction score
- 526
- Points
- 113
How innovative video generation technology will affect the reliability of content on the web.
Google made a significant leap in the field of video creation using artificial intelligence. The model is capable of creating high-quality realistic or surreal video clips lasting up to five seconds. It is able to animate static images or parts of them, flexibly responding to text requests from users.
One of the unique features of Lumiere is the ability to copy the image style and create a series of videos based on it on different topics, while maintaining the uniformity of the style. This feature can also be used to create branded content.
Lumiere can also transform the styles of the source video by turning objects into lego, origami, or flowers, depending on the parameters you set.
Especially noteworthy is the "inpainting" function, which allows you to remove unwanted elements from the video. The user can cover up an unnecessary part of the image, and Lumiere will automatically fill this space so that changes will not be noticeable.
The creators of Lumiere claim that the architecture of the "Space-Time U-net" model allows you to build a video in one pass, unlike previous models that created the start and end frames, and then tried to recreate what happens in between.
At the moment, Lumiere is a research project, which saves Google from having to restrict the system within the framework of copyright, misinformation, security, hatred, nudity, privacy, etc. Hypothetically, it can be used to create high-quality video deepfakes on absolutely any topic.
However, the model is now in closed access, which means that possible risks are minimized.
What is already clear is that Lumiere represents a huge leap forward in the field of neural network video generation. Enthusiasts are eagerly waiting for the model to be "felt" with their own hands and compared with existing analogues from other developers.
Подробнее: https://www.securitylab.ru/news/545511.php
Google made a significant leap in the field of video creation using artificial intelligence. The model is capable of creating high-quality realistic or surreal video clips lasting up to five seconds. It is able to animate static images or parts of them, flexibly responding to text requests from users.
One of the unique features of Lumiere is the ability to copy the image style and create a series of videos based on it on different topics, while maintaining the uniformity of the style. This feature can also be used to create branded content.
Lumiere can also transform the styles of the source video by turning objects into lego, origami, or flowers, depending on the parameters you set.
Especially noteworthy is the "inpainting" function, which allows you to remove unwanted elements from the video. The user can cover up an unnecessary part of the image, and Lumiere will automatically fill this space so that changes will not be noticeable.
The creators of Lumiere claim that the architecture of the "Space-Time U-net" model allows you to build a video in one pass, unlike previous models that created the start and end frames, and then tried to recreate what happens in between.
At the moment, Lumiere is a research project, which saves Google from having to restrict the system within the framework of copyright, misinformation, security, hatred, nudity, privacy, etc. Hypothetically, it can be used to create high-quality video deepfakes on absolutely any topic.
However, the model is now in closed access, which means that possible risks are minimized.
What is already clear is that Lumiere represents a huge leap forward in the field of neural network video generation. Enthusiasts are eagerly waiting for the model to be "felt" with their own hands and compared with existing analogues from other developers.
Подробнее: https://www.securitylab.ru/news/545511.php
