This blog post was originally published at NVIDIA’s website. It is reprinted here with the permission of NVIDIA.
The NVIDIA Maxine AI developer platform is a suite of NVIDIA NIM microservices, cloud-accelerated microservices, and SDKs that offer state-of-the-art features for enhancing real-time video and audio. NVIDIA partners use Maxine features to create better virtual interaction experiences and improve human connections with their applications.
Making and maintaining eye contact are rare in virtual settings because it is often difficult to align your gaze with the camera while holding a meeting or producing a video. Distractions, scripts, notes off to the side, and other factors add to the challenge of keeping eye contact.
Maxine Eye Contact solves this problem by aligning users’ gaze with the camera to simulate eye contact and increase engagement and connection. For more information, see NVIDIA Maxine Elevates Video Conferencing in the Cloud.
Flexible options with Maxine
There are several options, outlined later in this post, for integrating Maxine features into applications. Texel is an AI platform that provides cloud-native APIs to help you scale and optimize workflows for image and video processing. It makes integration easier and more cost-efficient for smaller developers using the cloud.
Co-founders Rahul Sheth, Texel CEO, and Eli Semory, Texel CTO, are enablers of scale. According to Sheth, “The Maxine Eye Contact SDK from NVIDIA is a state-of-the-art model, but the smaller shops and developers may have challenges in adopting the feature at scale out of the box. Our video pipeline API makes it easy to use any number of models on video in a highly optimized and cost-effective fashion.”
The collaboration with NVIDIA has saved Texel’s customers valuable development time and made this technology accessible to a broader range of developers and users.
Automatic Eye Contact with the Texel API
In the above video, see how AI enhances video communication by naturally redirecting the eyes toward the camera to enhance engagement and output.
Benefits of integration
Maxine allows for flexible, fast deployment and updates across any platform.
NVIDIA NIM Microservices
Here are the benefits of using NVIDIA NIM microservices to integrate Maxine features into your applications:
- Efficiently scales your applications, ensuring optimal performance and resource usage.
- Integrates easily with Kubernetes platforms.
- Enables deploying NVIDIA Triton at scale.
- Provides one-click deployment options, such as for NVIDIA Triton Inference Server.
NVIDIA SDKs
Here are the benefits of using NVIDIA SDKs to integrate Maxine features into your applications:
- Includes NVIDIA Triton Inference Server support for scalable AI model deployment.
- Enables seamless scaling across different cloud environments.
- Supports multi-stream scaling with improved throughput.
- Standardizes model deployment and execution across every workload to simplify AI infrastructure.
- Offers concurrent model execution, maximizing GPU utilization and throughput.
- Provides dynamic batching to improve inference performance.
- Enables model ensembles and business logic scripting for complex AI pipelines.
- Supports cloud, data center, and edge deployments.
Figure 1. NVIDIA Triton support enhances the throughput of concurrent streams
Using the Maxine SDK on Triton Inference Server provides a comprehensive, optimized, and scalable solution for AI model deployment and inference, using multi-framework support, advanced features, and robust model management capabilities.
Texel for simplified scaling
One of the key strategies for scaling with Maxine is through the integration and optimization of implementers, like Texel, which enable the service to cater to millions of users.
- Simplified API integration: Control the features you want to activate without having to manage complex backend processes. This simplifies the integration of Maxine features into applications.
- End-to-end pipeline optimization: Focus on using the features rather than infrastructure. The entire pipeline is optimized from input to inference to output, streamlining data flow and scaling behind the scenes.
- Custom model optimization: Bring your own custom models to optimize, reducing inference time and NVIDIA GPU memory usage. This makes it easier to scale custom AI solutions built on Maxine.
- Hardware abstraction: Use the latest NVIDIA GPUs without having to become a hardware expert, lowering the barrier to adopting advanced hardware acceleration.
- Efficient resource utilization: Run on as few GPUs as possible, potentially reducing costs and making scaling more economical.
- Real-time performance: Build more responsive applications that can edit images and videos with AI in real time.
- Flexible deployment: Choose hosted or on-premise deployment options for the scaling approach that best fits your needs.
The Texel team’s background in running large GPU fleets at scale (for example, at Snapchat) informs their approach to making NVIDIA-accelerated AI more accessible and scalable. You can effectively address the complexities of scaling and handle the demands of millions of users with the flexible integration options and the help of Texel.
Conclusion
The NVIDIA Maxine AI developer platform, coupled with Texel’s scalable integration solutions, offers you a powerful toolkit for creating cutting-edge video applications with advanced AI features. By using the flexible integration options provided by NVIDIA, you can efficiently implement sophisticated capabilities, like Eye Contact, for real-time enhancements.
The seamless scalability offered by Texel enables applications to grow from prototype to production-ready systems capable of serving millions of users. Leave the complexities of AI deployment and scaling to experts, and focus on building unique user experiences.
From enhancing day-to-day video conferencing to integrating AI technology, NVIDIA Maxine offers high-quality video communications for all professionals.
The latest Maxine production release is included exclusively with NVIDIA AI Enterprise, which enables you to tap into production-ready features such as NVIDIA Triton Inference Server, enterprise support, and more.
If you’re interested in early access, with non-production access to production and soon-to-be-released features, see the Maxine Early Access program.
To help improve features in upcoming releases, provide feedback on the NVIDIA Maxine and NVIDIA Broadcast App survey.
For more information, see Texel’s video APIs and contact the Sales team with any questions.
Related resources
- GTC session: Using AI to Enhance Immersion and Communication in Video Conferencing
- GTC session: Driving Innovation: Generative AI and Industrial Digitalization in the Automotive Industry
- GTC session: Global Innovators: Scaling Innovation With NVIDIA AI
- SDK: Video Effects SDK
- SDK: Video Effects Microservice
- SDK: Audio Effects Microservice
Trisha Tripathi
Product Manager, NVIDIA
Maryam Motamedi
Product Manager, NVIDIA
