At CES 2025, Nvidia showed off a prototype AI avatar that lives on your PC’s screen. R2X is an AI assistant that looks like a figure from a video game. It can help you find your way around in computer apps.
The Nvidia AI models are used to render and animate the R2X image. Users can run the avatar on any popular LLM, like OpenAI’s GPT-4o or xAI’s Grok. People can text or talk to R2X, send it files to be processed, or even let the AI assistant see what’s going on your screen or camera in real time.
Recently, tech companies have been making a lot of AI characters, not just for video games but also for business and consumer users. Some people think these avatars could be a good way for AI helpers to interact with people, even though the early demos are weird. Nvidia’s R2X tries to make an AI helper that looks and feels like a person by combining advanced LLMs with features found in generative video games.
In the first half of 2025, the company wants to make these models available to everyone. Nvidia thinks of this as a fresh way for developers to make apps, since users can connect their favourite AI software and even run these models locally.
Like Microsoft’s Recall feature, which has been pushed back because of privacy concerns, R2X can constantly take screenshots of your screen and send them to an AI model to be processed. By default, this feature is turned off. When it’s turned on, it can give you feedback on programs that are running on your computer and help you with things like hard coding tasks.
You can still try out R2X, and even Nvidia agrees that it has some bugs that need to be fixed. During tests with Parhlo World, Nvidia’s image had an uncanny valley feel to it. Its face got stuck in strange places, and its tone could be a little hostile at times. Also, having a humanoid image stare at me while I work seems a little strange to me.
R2X usually gave good directions and saw what was on the screen correctly. And then the avatar stopped being able to see the screen at all. At one point, it gave us the wrong directions. This could be a problem with the AI model itself (in this case, GPT-4o), but the example shows what this early technology can’t do.
An Nvidia product lead showed how R2X can see the apps on your screen and help users with them in one example. In particular, R2X helped us use the generative fill tool in Adobe Photoshop. The picture we chose shows Nvidia CEO Jensen Huang standing with two restaurant workers in an Asian restaurant. When Nvidia’s character had a bad dream, it told people the wrong way to find the generative fill feature in Photoshop. Later, it lost the ability to see the screen, but the character was able to see the screen again after we changed the AI model to xAI’s Grok.
This time, R2X could take a PDF from the PC and then answer questions about it. A local retrieval augmented generation (RAG) feature powers this process. This feature lets these AI avatars get information from a document and use the base LLM to process it.
This is how the images look thanks to AI models from Nvidia’s video game business. Nvidia’s RTX neural faces program is used to make avatars. A new model from Nvidia called Audio2FaceTM-3D is being used to control the movement of the face, lips, and tongue. At times, that model seemed to stop working, which put the avatars’ faces in strange situations.
Another thing the company says is that these R2X robots will be able to join Microsoft Teams meetings and work as personal assistants.
It is the goal of an Nvidia product lead to give these AI avatars agentic skills as well. This means that R2X could one day do things on your desktop. It looks like these skills will not be available for a while, and to get them, companies like Microsoft and Adobe, which are working on similar agentic systems, would probably need to work together.
Also Read: Nvidia Reveals Rtx 5000 Series at Ces 2025
At first glance, it’s not clear how Nvidia is making the voices in these goods. When R2X uses GPT-4o, its voice sounds different from all of ChatGPT’s pre-set voices. On the other hand, xAI’s Grok robot doesn’t have a voice mode yet.
What do you say about this story? Visit Parhlo World For more.
