Topaz Labs, known for his photo and video software, has developed a very good artificial intelligence model that improves the quality of videos.
The diffusion model, called Project Starlight, focuses directly on video restoration and makes a first in the field according to the company. Very low quality videos renewing and clarifying and increasing the resolution model, According to the examples shown, it shows an impressive performance. In a statement about the model, which can also develop new images taken with professional cameras, Topaz conveys: “Our diffusion -based approach, which is created with a completely new model architecture, uses more than 6 billion parameters and benefits from the latest Nvidia GPUs. This sets a new standard for the most dynamic and comprehensive video improvement method and artificial intelligence -based video restoration we have created so far. ” You can watch the promotional video shared for Project Starlight, which will be opened early:
https://www.youtube.com/watch?v=ppI1dc-ogu
Fully Blind Restaurant: Our Model Automatically Denoises, Deblurs, Upscales, Anti-Aliases. Simply Feed in Your Footage and Watch it Transform! No manual tweaks, no parameter tuning.
🎥 Our Model Can Restore Footage Shot on Professional Cameras. This Example 👇 Was Filmed Using… pic.twitter.com/tmlpz0jm0t
– Topaz Labs (@topazlabs) February 6, 2025
You may be interested in
Before that, Omnihuman-1, which was developed by ByTedance, attracted attention. Omnihuman-1as you can see in the following videos, it can produce realistic content that cannot be distinguished. In order to produce these videos, a single reference photo and sound clip is sufficient, while certain arrangements can be made on videos created with written commands. That he can make changes to real videos (For example, he can change the mouth movements of the person in the video) The system still does not work flawlessly, but it can produce realistic videos that cannot be distinguished when it is not very careful.
https://www.youtube.com/watch?v=n6hkcs2pj0q
https://www.youtube.com/watch?v=z8pxnogc9w
Google Veo has been the new version of this area. Developed by Google Deepmind, Veo can create realistic videos from the OpenAI Service Sora, which has been opened to everyone in recent months. The new version and 4096 x 2160 pixel resolution, which does not offer videos of two minutes, starts Deepmind, the new model that can create clips in various styles He states that he has achieved better results in physics calculations and camera controls.. Veo 2, which can produce much clearer images compared to the first version, is a better than the company. “cinematography” has an understanding. In this way, when commanding the model, it can refer to a specific type of film, cinematic effect or lens. The system understands them and prepares the video to be produced accordingly.
In addition to Veo 2, which is said to be better in human movements Improvements were also made for Imagen 3 model. The artificial intelligence system, which produces visuals from Google, can now give more bright and realistic results. According to the statement, Veo 2 will be presented to Google Labs users in the USA gradually. Test users will be able to produce images of up to eight seconds of 720p resolution. Imagen 3 can currently be used by Google Labs users in more than 100 countries via ImageFX.