HomeWinBuzzer NewsHow NVIDIA Graphics Research is Pushing the Boundaries of Generative AI

How NVIDIA Graphics Research is Pushing the Boundaries of Generative AI

NVIDIA says it is working with universities and researchers to enhance the capabilities of generative AI to offer more control and customizations.


has written a blog post describing the advancements it is making in . According to the company, NVIDIA Graphics Research is at the forefront of advancing generative AI and neural graphics, which combine deep learning and computer graphics techniques.

To highlight its progress, NVIDIA Graphics Research will present around 20 research papers at SIGGRAPH 2023, the computer graphics conference that will be held between August 6 and 10 in Los Angeles. The papers showcase the latest innovations in generative AI models, inverse rendering tools, neural physics models and neural rendering models.

Generative AI is a branch of artificial intelligence that can create new content from scratch, such as images, text, audio and video. Generative AI has many applications in fields like art, design, entertainment, and education.

However, generative AI also poses many challenges, such as how to make the generated content realistic, diverse, personalized and interactive.

Generative AI Models: From Text to Personalized Images

NVIDIA says one of the most exciting applications of generative AI is to turn text into images. This can enable creators to quickly generate concept art or storyboards for films, video games and 3D virtual worlds. For example, a text prompt like “children's toys” can produce images of stuffed animals, blocks or puzzles.

We have already seen this sort of text-to-image capability with . Announced in March, Bing Image Creator uses OpenAI's DALL-E image generating AI and combines it with Bing Chat. Users can provide a natural language text prompt and receive an image response from Bing.

NVIDIA says it is working on providing more control over the output of the generative AI model. For instance, a creative director for a toy brand may want to visualize a specific teddy bear in different situations, such as a teddy bear tea party. To achieve this level of customization, researchers from Tel Aviv University and NVIDIA have developed two techniques that allow users to provide image examples that the model learns from.

Inverse Rendering Tools: From Images to 3D Objects

Another challenge in generative AI is to transform 2D images into 3D objects. This can help creators to create realistic and detailed 3D models from photos or sketches. For example, an architect may want to convert a floor plan into a 3D building.

To solve this problem, researchers from NVIDIA and several universities have developed inverse rendering tools that use deep learning to infer the shape, texture, lighting and material properties of 3D objects from 2D images.

Neural Rendering Models: From Simulation to Visualization

Another challenge in generative AI is to visualize the generated content with high quality and performance. This can help creators to render photorealistic images or videos for various purposes. For example, a filmmaker may want to render a scene with complex lighting effects.

To address this issue, researchers from NVIDIA and several universities have developed neural rendering models that use deep learning to enhance the visual quality and efficiency of rendering techniques.

Microsoft Positions Itself to Compete with NVIDIA

While NVIDIA is pushing ahead with and development, is working to compete with the compay. This week, we reported on Microsoft partnering with AMD to develop AI chips that will challenge NVIDIA.

Microsoft's chip is known internally as “Athena” and will compete with Nvidia's Hopper H100 GPU, which is also used by Microsoft for its Azure cloud platform.

The Athena chip is expected to support large-scale natural language processing (NLP) applications, such as conversational agents and text summarization. Microsoft wants its AI CPU to reduce how much the company relies on Nvidia and also deliver a more custom experience to users. Athena is not expected to be announced until 2024.

Tip of the day: To prevent attackers from capturing your password, Secure Sign-in asks the user to perform a physical action that activates the sign-in screen. In some cases, this is a dedicated “Windows Security” button, but the most common case in Windows is the Ctrl+Alt Del hotkey. In our tutorial, we show you how to activate this feature.

Luke Jones
Luke Jones
Luke has been writing about all things tech for more than five years. He is following Microsoft closely to bring you the latest news about Windows, Office, Azure, Skype, HoloLens and all the rest of their products.

Recent News