Leading Generative AI firm, Yepic.AI, chooses OVHcloud's high-end GPUs to deliver its innovative solutions
OVHcloud and Yepic.AI
Around 30
GPU instances in production.
1 million videos
generated by the Yepic Studio service.
Monthly income
from $2k to $50k in five months.
Executive summary
A robotic receptionist greets visitors arriving at a campus or a company; an ultra-realistic presenter reads out a firm’s quarterly results on its YouTube channel; whilst your face is deepfaked and your voice dubbed so you can speak to someone on the other side of the world. This is the revolutionary AI video generator technology that startup - Yepic.AI - is offering.
AI video generator technology is a form of Generative AI, which works by generating content (such as videos, text or images) in response to user prompts. It is already familiar in the form of deepfakes, but the founders of Yepic.AI are determined to make it not only a useful tool for companies, but a new means of communication that could truly revolutionise international trade.
In 2021, the startup launched its first AI video generator solution, the Yepic Studio online service. This tool allows users to generate a video based on a simple text. Simply choose an avatar and type what you want to say. The video becomes available a few minutes later. And the results are stunning. They have acquired 500 devout monthly users in the last 5 months. The company's second AI generator tool is a simultaneous voice translation service known as VidVoice. Users can have conversations with anyone in the world and receive a simultaneous translation without the need for human interpreters. And you don’t even need a transcription of what participants are saying.
Aaron Jones, the CEO of Yepic.AI, explains how these ground-breaking AI video generator tools work: “We are developing tools to help companies communicate by video, but it really is more than just for business usage. Our goal is to optimise how people communicate in global organisations and make it easier to access healthcare. Twenty-five million Americans don’t speak English, and in the UK nearly 1 million people can’t speak English well enough to access healthcare without an interpreter. With a tool like this one, the whole world will gain access to knowledge and services in their mother tongue, simply by speaking.”
The challenge
Yepic.AI’s core expertise is Generative AI, specifically AI video generators. The startup works closely with the field’s most advanced researchers in the United Kingdom, including those at the Centre for Vision, Speech and Signal Processing (CVSSP) as well as other researchers from the universities of Oxford, Warwick, Birmingham, and Kingston University. This form of Generative AI requires a huge amount of machine power to generate perfectly natural-sounding videos that are synchronised with text readings. Yepic.AI's video generator technology is based on Machine Learning algorithms and generates a video in real-time, just like the VidVoice service. This process requires the most efficient GPUs on the market with the maximum amount of possible memory.
But Yepic.AI is still just a ten-person startup. Its human resources are nothing like those available to Generative AI giants like as Facebook, Microsoft, and Deepmind. As a result, its researchers needed extremely powerful, as well as highly industrialised, tools to create, test, and produce their Generative AI algorithms. They also had to find a reliable and efficient GPU infrastructure. That’s why Yepic.AI turned to OVHcloud.
The solution
While Yepic.AI researchers began developing their AI video generator tools using Cloud services provided by American hyperscalers, they very quickly switched to OVHcloud services. Aaron Jones explains why: “I found out about OVHcloud at a virtual event. I felt that we shared the same values and they quickly invited us to join their startup program.” The founder of Yepic.AI was on the lookout for a solid partner for the scale-up phase of Yepic Studio. “A business can use this tool to enter a script in forty languages, choose an avatar and generate a video in just a few minutes. In these early stages, the service is available online to business users.” This AI video generator tool consumes a lot of GPU resources and if the launch was to be successful – a must for any startup – it meant having an efficient and reliable infrastructure.
Yepic.AI uses around thirty OVHcloud ML Serving GPU instances to provide its Generative AI services. Meanwhile, researchers are making good use of Jupyter notebooks in managed mode using OVHcloud’s AI Notebooks solution. “Being able to use OVHcloud notebooks has been a game changer for us. The seamless integration between AI Notebooks and ML Serving saves us a lot of time. We no longer have to go through weeks of waiting before bringing in an innovation: we can automate how we load an algorithm on multiple GPUs from the notebook.” The agility and speed of execution allows the startup to bring in new features very quickly: a key asset for a fledgling business, especially one operating within a fast-moving space such as Generative AI.
The result
Even though Yepic.AI’s technical infrastructure is starting to implement OVHcloud’s managed databases, the startup’s core activity is based on the GPUs that are made available by the European cloud leader. “OVHcloud’s AI Notebooks give us access to much more GPU memory than we could get from other cloud providers. For a company whose business is 100% machine learning, that’s absolutely crucial.” Unlike the main cloud providers on the market, OVHcloud provides access to NVidia’s Tesla V100S GPU rather than the classic V100 version. The V100S version offers 32 GB of memory, which is double the memory offered by the V100 from other cloud providers. “Having 32 GB of memory rather than 16 GB makes all the difference when it comes to outputting video at 60 frames per second. To achieve this level of performance, all the data must be stored in GPU memory. That level of performance and rendering quality is impossible if you have less memory and need to write data to disk.”
Likewise, the integration between OVHcloud AI Notebooks and OVHcloud ML Serving is a key asset for the startup: “Our researchers can try out new AI models in a test environment very quickly. The Notebook architecture provided by OVHcloud allows us to make fast adjustments to the resources consumed, which helps save on operating costs while bringing a huge productivity gain for our team.” The founder of Yepic.AI estimates that productivity gain to be multiplied by a factor of 3 to 4 because by using a service handled by OVHcloud, it is much simpler to manage the different versions of the CUDA, TensorFlow and PyTorch libraries. “Managing framework versions is a real problem. Look at the example of PyTorch, each new version brings its share of differences and incompatibilities. You need to be able to run the versions of PyTorch with which your code was written. With the concept of containers and OVHcloud’s AI Notebooks architecture, it is relatively easy. Having the right containers is a key point in delivering services to our customers without delay.”
“We’ve tried out all the leading American hyperscalers and OVHcloud is certainlythe Cloud service we found most user friendly. In just a few days, the whole teamwas fully operational, so our researchers could start innovating more quickly. Theproduction infrastructure delivered by OVHcloud brings us the best performing GPUs,which is absolutely critical for our business.”Aaron Jones, CEO of Yepic.AI