At Higgsfield, we built the leading text to video platform for social media. 

We’re excited to introduce NOVA-1, our advanced text-to-video world model that gives precise control over the look and feel of a piece of content.

NOVA-1 transforms social media content production by enabling premium looking video at non-premium costs. None of this would be possible without our partnership with Nebius and integrating into their cloud GPU network for L40S inference.

The Key Benefits of NOVA-1 are: 

1. The ability to train a custom version of NOVA-1 that generates consistently on-brand videos.

2. High prompt coherence and control over generations. From adjusting an existing video to creating a highly imaginative scene from scratch, our model relies on the network of L40S for generation.

3. Highly dynamic and fluid human characters with varied motion. Imagine generating lifelike human characters that can be controlled to move in a precise manner. Our technology makes this possible without having to shoot on location or set up a live production.

With Nebius, we’re able to efficiently utilize clusters of L40S GPUs for video model inference. It is incredibly efficient - we see 40% cost efficiency gains with L40s without sacrificing content quality or video generation speed.  We also run a lot of different models to power our tech. With these requirements, we need multiple solutions to keep our network healthy:

1. Managed Kubernetes to efficiently manage and scale a lot of VMs to handle increasing load

2. Utilize Networking resources for infrastructure security 

3. Cost-effective Object Storage for high-volume data management 

4. Ability to manage infrastructure with Terraform via Nebius

For the currently 400k+ users of our video ai entertainment app, Diffuse, our video model gives users an opportunity to generate delightful and share-worthy personalized videos with just one selfie. This had led to a highly retentive experience for users (50% action rate of sharing or downloading generated content) and establishes us as the leading video generation model on the market today.  And we’re able to support paying users with faster inference using the network of L40s.

We are happy to support such ambitious AI startups as Higgsfield with infrastructure for training and inference. Being an AI-centric cloud, we see it as our mission to be a partner for companies who use AI to develop disruptive innovation across industries, and Higgsfield is definitely one of them. Higgsfield is a team of authentic creators who always seek out new solutions in order to be more efficient and develop faster. This resonates with our mindset and helps us to evolve as well.

Andrey Korolenko, Chief Product Officer, Nebius AI

We’re thrilled about the early wins we achieved with Nebius in this short time and eager to see how our partnership will push the boundaries of video generation.

Onwards and Upwards!