In today’s fast-paced digital landscape, captivating video content has become the driving force behind successful brand engagement. From product explainers to targeted marketing campaigns, videos offer an immersive and effective way to convey messages. However, for startups, producing high-quality videos can often be a daunting task due to limited resources and technical expertise. Enter Hour One, a startup that has harnessed the power of Microsoft Azure to revolutionize video content creation. Hour One is part of Microsoft for Startups Founders Hub and has heavily leveraged the technology benefits from the program to improve their product, so we sat down with Hour One to talk about how Microsoft technology helped them address critical challenges in the video production process.
Hour One’s video revolution
Hour One has developed a platform called Reals that empowers businesses to create professional videos featuring lifelike virtual humans. The goal is to simplify the video creation process, making it accessible to all, regardless of technical expertise. Reals leverages artificial intelligence to generate compelling video scripts, enabling users to effortlessly transform ideas into engaging visual stories.
In a competitive digital landscape, driving audience engagement through videos can be an uphill battle. Creating videos that resonate with viewers requires a deep understanding of storytelling, design, and production techniques. Hour One recognized this challenge and set out to provide an innovative solution that empowers startups and businesses to create impactful videos without the need for extensive resources or technical know-how.
Using AI to simplify the video creation process
One of the biggest barriers to video content creation is the “blank canvas effect,” where users struggle to start from scratch. Hour One’s vision was to provide an AI copilot that simplifies the video creation process, allowing users to generate scripts with a single sentence or prompt. By integrating Azure OpenAI Service, Hour One’s Reals platform generates initial text for videos based on user prompts, dramatically reducing the time and effort required to create compelling scripts.
When building the Reals platform, they wanted users to be able to use a single sentence or detailed prompt to create complete scaffold of a video, dramatically reducing the time and effort required to create compelling scripts. Reals also needed to provide customizable backgrounds and other elements, ready for tweaking. Beyond this, the platform would have to allow for titles, images, and text to be effortlessly fine-tuned and then render the final video quickly.
Other technical challenges included:
- All speech in the video had to be as natural as possible.
- Speed and performance were a must to produce immediate results and to iterate quickly even under demand conditions.
- Content had to be dynamic so users could view and update the video before publishing. (If the editing process took days to complete, the platform would not be usable.)
- Performance could not degrade, meaning the platform had to achieve the same immediate results even as the number of users scaled sky-high.
Azure Services Used by Hour One
When evaluating the technological landscape and determining which provider would best fit their needs, Hour One landed on using a range of Azure services for their Reals platform.
Using Azure OpenAI Service to generate video scripts based on user prompts
Azure OpenAI Service provides access to one of the world’s most advanced generative AI models combined with Azure AI-optimized infrastructure, enterprise-readiness, compliance, data security, and privacy controls, along with many integrations with other Azure services. The Reals platform uses Azure OpenAI Service to create prefilled scripts with multiple scenes and compelling text based on a simple one-line prompt from the user, helping to produce high-quality content quickly and easily.
Using Azure Kubernetes Service (AKS) for scaling and performance
Scaling is a crucial consideration for startups aiming to meet growing demand. Azure Kubernetes Service (AKS) simplifies the deployment, management, and scaling of containerized applications on Azure, which allowed Hour One to dynamically allocate GPU resources, ensuring smooth scaling to handle both baseline and peak workloads. They used AKS to drive the inference cluster at the heart of its Reals platform, helping users create interactive content fast.
AKS supports GPU-enabled Linux node pools, which means it can run compute-intensive Kubernetes workloads via cutting-edge NVIDIA GPUs. This scalability was essential to support Hour One’s user base as it expanded, without compromising on performance.
Using Azure Container Registry (ACR) for security
This secure, scalable, and private Docker registry hosted on Azure allows Hour One to store and manage container images. Azure Container Registry is integrated with Kubernetes, letting Hour One easily deploy and scale the Reals platform.
Deploying the Hour One solution
According to Hour One CTO Lior Hakim, “Azure’s scalability and reliability have allowed us to process vast amounts of data, train complex algorithms, and optimize our model’s performance, ultimately providing our customers with a superior video creation experience.”
Here’s a brief overview of the video creation process using the Reals platform:
- User input from the Reals web client or a mobile client passes through the API to the AKS infrastructure powering the platform.
- Azure OpenAI generates initial text for the user’s video based on a prompt provided by the user or fills in text on a video template.
- This text is passed back to AKS, which renders the video, matching up expressions and facial movements with the generated text.
- The user can then tweak the video, selecting a different virtual human and modifying the text, background, and many other features, until they are completely satisfied.
- The Reals platform uses a proprietary microservices architecture to render the final high-quality video.
Here’s what the process looks like at a schematic level:
Hour One’s actual transition to using Azure components took place in two phases:
- Phase 1. Development Cluster: Hour One’s team found it unbelievably simple to install the inference cluster within their CI/CD process, as well as to seamlessly integrate AKS within the container registry:
- Transfer of research infrastructure to Azure (2 months)
- Development and iteration, training of the foundational model on the Azure cluster, continuous training of the next generation of the model (1.5 years)
- Phase 2. Production Cluster: This phase involved shifting actual production workloads to Azure. According to Hour One CTO Lior Hakim, “Everything just worked smoothly.”
- Transfer of production workloads, up and running in Azure (1 month)
How Hour One benefits from Azure
The Hour One team has benefited from Microsoft Founders Hub in several key ways:
- Ease of Integration: Perhaps the greatest benefit Hour One experienced was the seamless transition to and integration with Azure, including a large volume of heavy-duty workloads. Thanks especially to support from the Azure team, the entire process was seamless.
- Cloud credits to train a foundational model with efficiency: Through Founders Hub, Hour One received cloud credits towards a variety of Azure services, giving them access to cutting-edge infrastructure and resources
- Production-Ready AI: Azure OpenAI provides a production-ready, leading-edge AI model, but Hour One can still benefit from the OpenAI playground environment during development before transitioning to Azure OpenAI for production capabilities. Azure OpenAI also scales beautifully, so that Hour One can use compute resources more efficiently and handle larger workloads.
- 2.2x Faster Processing: Access to Azure OpenAI Serive has also given Hour One a 2.2x boost in processing performance, even for the most demanding tasks. Advanced Microsoft Azure infrastructure components, such as NVIDIA GPUs, combine with Azure OpenAI for enhanced processing power and data throughput, all of which contributes to reduced latency.
- Faster Scaling: Using efficient inference machines and leveraging Microsoft Azure infrastructure, such as AKS and ACR, Hour One has optimized resource usage while cutting the costs of both training and inference. Thanks to AKS’ scalability and NVIDIA GPU provisioning, Hour One has dramatically expanded its service capabilities. Being able to dynamically allocate and provision GPU resources in response to workload demands means that Reals can handle both baseline and peak workloads while upholding superior service quality. Furthermore, machine scaling time has been cut by tens of seconds relative to other providers, making the system even more responsive and efficient. Plus, access to these GPUs and the ability to scale up or down without incurring CapEx have had major business impacts.
- Network Reliability: Microsoft Azure has provided Hour One with superior network throughput; even transferring data from other clusters to Azure is very fast.
“Being able to efficiently scale up or down quickly eliminates the need for large upfront hardware investment, thereby not only reducing overall expenditure but also increasing agility and turnover speed.” – Lior Hakim, CTO at Hour One
Using Azure to empower video content creation for everyone
Hour One has been able to rapidly iterate and improve the quality of the virtual humans provided by the Reals platform thanks to industry-leading Azure components and support from Microsoft, and we’re so excited to be helping Hour One democratize video creation.
“Microsoft’s substantial cloud offering and the cloud credits we received through the Microsoft for Startups Founders Hub played a pivotal role in empowering us to train our foundational virtual human model with unprecedented efficiency. With access to cutting-edge infrastructure and resources, we harnessed the power of Microsoft’s cloud services to achieve the best lip-sync and visual quality for our virtual humans.” – Lior Hakim, Co-founder and CTO
Just as Hour One harnessed Azure’s power to revolutionize video content creation, any startup can leverage Microsoft’s technology to overcome obstacles, scale with ease, and create meaningful connections with your audience.
Microsoft for Startups Founders Hub members receive Azure cloud credits that can be used towards a variety of Azure services, including Azure OpenAI Service, as well as expert advisory from the team at Microsoft to build and deploy their solutions. Founders Hub is open to all startups and takes just minutes to apply – sign up today.
