How ProSieben scaled Germany’s Next Top Model TOPSHOT for real-time AI portraits with Gcore
- October 1, 2025
- 2 min read
To celebrate the 20th season of Germany’s Next Top Model (GNTM), ProSieben’s marketing team launched GNTM TOPSHOT, an AI-powered feature in the Joyn app that instantly transforms user photos into studio-grade, show-inspired portraits.
At broadcast scale, the challenge was steep: handle massive primetime spikes, deliver results instantly, and guarantee strict EU privacy compliance. To make it possible, ProSieben turned to Gcore Everywhere Inference.
“We needed something that could scale on demand, deliver in seconds, and keep data local. Gcore’s infrastructure let us bring a creative idea to life - without breaking the user experience.”
- Simon Hawe, Technical Lead, Joyn
The challenge: broadcast-scale creativity in real time
TOPSHOT had to satisfy five tough, non-negotiable requirements:
- Primetime surges. Usage surged before, during, and after live broadcasts across Germany, Austria, and Switzerland.
- Ultra-low latency. Fans expected results within 5 - 10 seconds per portrait, end-to-end.
- Privacy by design. Joyn deletes all photos immediately to keep user data truly private - no caching, no storage. Caching wasn’t an option; every request had to be generated fresh.
- High fidelity. Each portrait required a ~100-stage pipeline for segmentation, relighting, skin/pose preservation, and compositing, to match GNTM’s signature aesthetic.
- Frictionless UX. No login required. The feature had to “just work,” even on mobile connections.
Turning fans into models in real time
ProSieben needed an inference platform that was scalable, privacy-compliant, and easy to integrate. Gcore Everywhere Inference delivered:
- One endpoint, nearest node. Smart Routing automatically sent each request to the closest GPU endpoint, minimizing jitter and wait times.
- Autoscaling GPUs. Serverless orchestration spun GPU capacity up or down in real time, handling unpredictable primetime peaks.
- EU-ready deployments. Hybrid support (on-prem, Gcore Cloud, public cloud) gave ProSieben full flexibility on data residency.
- Optimized for image workloads. Everywhere Inference ran TOPSHOT’s complex pipelines on NVIDIA H100, A100, and L40S GPUs - excellent for generative image models.
“Our biggest challenge was combining visual fidelity with real-time response. Gcore’s Smart Routing and auto-scaling made that possible at primetime scale.”
- Benjamin Risom, Chief Product Officer, Joyn
The architecture allowed ProSieben to:
- Route all traffic through a single inference endpoint, fronted by their own load balancer
- Keep portrait generation under 10 seconds - even during broadcast surges
- Meet strict privacy guarantees: no logins, no storage of inputs or outputs
- Deliver a seamless experience inside the Joyn app
TOPSHOT went live with five portrait scenes in April 2025, with three more added weeks later.
“We could focus on the creative, knowing the infrastructure would scale with us. That made it possible to deliver something really special for our viewers.”
— Sebastian v. Wyschetzki, Creative Lead, Seven.One Entertainment Group
Real-time engagement, broadcast scale
TOPSHOT launched into a season already driving cultural buzz.
- The GNTM Season 20 finale (June 19, 2025) drew 3.87M viewers with a 22.4% share in the 14 - 49 demo.
- Joyn saw 10M viewers in April 2025 (+80% YoY) and a 40% increase in watch time in Q1 2025 YoY.
- TV Total host Sebastian Pufpaff demoed TOPSHOT live on air, praising the visuals and sparking organic uptake.
- Trade press highlighted the “scalable Gcore infrastructure” behind the feature.
“Gcore’s platform gave us regional performance, privacy control, and GPU scaling without the heavy lifting of building and managing infrastructure ourselves.”
— Paolo Garri, Infrastructure Architect, Joyn
What’s next?
Building on TOPSHOT’s success, ProSieben is playing with new potential fan-facing AI experiences: video portraits, real-time filters, or stylized animations. With Gcore’s flexible infrastructure, the team is free to keep experimenting without re-architecting.
“The success of TOPSHOT showed us what’s possible. Now we’re asking: how far can we take this?”
— Jutta Meyer, Executive VP Marketing & Creation, Seven.One Entertainment Group
More case studies
Higgsfield AI kickstarts partnership with Gcore for scalable AI infrastructure and Managed Kubernetes support
Founded in 2023, Higgsfield is building the Video Reasoning engine for the attention economy. Its AI-native, browser-based platform condenses ideation, editing, and post-production into a single workflow, enabling creators and enterprises to produce cinematic-quality short-form video in minutes instead of weeks.Higgsfield delivers fast, controllable, and scalable outcomes that preserve narrative continuity and cultural resonance across media, marketing, and brand communication. With operations spanning the US, Europe, and Asia, Higgsfield is headquartered in Silicon Valley and backed by world-class investors and veteran technologists with a track record of billion-scale products and outcomes.As they prepare for scale and increasing demand, Higgsfield needed robust, flexible infrastructure that could meet the needs of their dynamic workloads and rapidly growing user base.What it takes to power generative AI at scaleHiggsfield had worked with other GPU providers, but struggled with limited capacity and the lack of scalable orchestration options. The generative platform relies on running large-scale AI models efficiently, so their team's key infrastructure priorities were:Instant access to high-performance H100 GPUs with the ability to scale up and down based on project demandAutoscaling GPU infrastructure to support unpredictable, high-volume generative AI workloadsManaged Kubernetes with GPU worker nodes, load balancers, and cloud networking for ease of orchestration, autoscaling, and reliabilityFast onboarding and close support to move quickly from testing to deploymentTransparent and predictable pricing with fast and simple contracting, and PAYG or commitment models available.Availability and flexibility for future expansionWhy Gcore infrastructure stood out from the crowdHiggsfield approached Gcore in need of a large volume of H100 GPUs immediately, and with the flexibility to scale on demand. Gcore provided rapid access to the required H100 GPUs, helping Higgsfield eliminate supply constraints and meet fast-moving development timelines.Transparent pricing gave Higgsfield budget predictability and easier cost control, which was essential for their high-frequency release cycles. They also valued Gcore’s commitment to sustainable hardware design, high reliability and uptime, and 24/7 availability of DevOps engineering support.Additionally, deploying infrastructure through the Gcore Sines 3 cluster in Portugal provided the regional flexibility and high-performance Higgsfield needed to support its platform.Higgsfield chose Gcore for its ability to deliver managed Kubernetes with GPU worker nodes, enabling them to scale dynamically, flexing compute resources based on real-time user demand. Speed and flexibility are essential to Higgsfield's operations: They expect to start cooperating with partners quickly and scale capacity on demand. The streamlined service offering, fast onboarding, and highly responsive support that Gcore provides enabled them to do exactly that.“The combination of GPU scaling, H100 availability, and Managed Kubernetes was invaluable for us. Gcore gave us the control and flexibility our engineering team needed to move flexibly and fast.”— Alex Mashrabov, CEO, Higgsfield AIA fast, hands-on start with dedicated engineering supportGcore’s team provided dedicated engineering support and helped Higgsfield test their workloads through a one-week trial. After validating performance, Higgsfield quickly transitioned to full deployment.“Direct access to Gcore’s engineering team made the onboarding smooth and efficient. We could test and validate quickly, then scale up without friction.”— Anwar Omar, Lead Infrastructure Engineer, Higgsfield AIScalable performance and a strong foundation for growthWhile it’s early days, Higgsfield is already live and actively running GPU-powered workloads with Gcore in production.The key outcomes so far include:Seamless deployment to a managed Kubernetes environment with GPU worker nodes and autoscalingOn-demand access to H100 GPUs for compute-intensive generative workloadsKubernetes-based orchestration for efficient container scaling and resource optimizationScalable infrastructure that flexes based on demandA strong foundation for future product growth and global scalingWhat’s next?Higgsfield is currently exploring the possibility of extending the relationship beyond model training and into distributed inference infrastructure with Everywhere Inference.Their product roadmap involves releasing new features at a high velocity, often requiring larger GPU volumes for short periods—making flexible infrastructure a must. Gcore’s scalable, on-demand model supports this cadence without overprovisioning.We’re excited about the potential of our partnership with Gcore. The team has been incredibly responsive, and the infrastructure is a great fit for Higgsfield. We’re actively exploring additional possibilities—from Everywhere Inference to broader scaling—and we’re looking forward to seeing where this collaboration can take us next.— Alex Mashrabov, CEO, Higgsfield AI
Leonardo AI delivers high-speed, global content creation with Gcore AI services
Leonardo.Ai helps creators turn ideas into stunning AI-generated content in seconds. Headquartered in Australia and now part of Canva, the company gives game developers, designers, and marketers powerful tools to generate and refine images, videos, and creative assets in real time.As James Stewart, DevOps Engineering Manager at Leonardo.Ai explains, the team’s top priority is speed. Their north-star value is “go fast”, taking ideas to prototype and release at an impressive pace. But delivering that kind of speed at scale takes serious GPU infrastructure and deep levels of expertise around orchestration.Seeking speed, scale, and infrastructure maturity under pressureDelivering AI speed at scale for customers worldwide requires powerful, on-demand GPU inference infrastructure. Early on, Leonardo found that limited GPU availability and high cost were bottlenecks.GPUs make up a significant part of our operating costs, so competitive pricing and availability are crucial for us.James Stewart, DevOps Engineering Manager, Leonardo.AiWith big growth goals ahead, Leonardo needed an efficient, flexible GPU provider that would support their plans for speed and scale. They looked at AI providers from global hyperscalers to local GPU services. Some providers looked promising but had no availability. Others offered low prices or easy access (no long-term commitment) but were missing essential features like private networking, infrastructure-as-code, or 24/7 support.Cheap GPUs alone weren’t enough for us. We needed a mature platform with Terraform support, private networking, and reliable support. Otherwise, deployment and maintenance become really painful for our engineers at scale.James Stewart, DevOps Engineering Manager, Leonardo.AiFortunately, they found what they were looking for in Gcore: solid GPU availability thanks to its Northern Data partnership, a fully-featured cloud platform, and genuinely helpful technical support.We chose Gcore for overall platform integration, features, and support. Compared to some of the less capable GPU providers we’ve utilized, when using Gcore our engineers don’t need to battle with manual infrastructure deployment or performance issues. Which means they can focus on the part of the job that they love: actually building.James Stewart, DevOps Engineering Manager, Leonardo.AiFinding a flexible provider that can meet Leonardo’s need for speedLeonardo AI needed infrastructure that wouldn’t slow innovation or momentum. With Gcore, it found a fast, flexible, and reliable AI platform able to match its speed of development and ambition. Leonardo chose to run their inference on Gcore GPU Cloud with Bare Metal hosting, offering isolation, power, and flexibility for their AI workloads. Their demanding inference workloads run on current-gen NVIDIA H100 and A100 GPUs with zero virtualization overhead. This means their image and video generation services deliver fast, high-res output with no lag or slowdowns, even under the heaviest loads.On-demand pricing lets Leonardo AI scale GPU usage based on traffic, product cycles, or model testing needs. There’s no overprovisioning or unnecessary spending. Leonardo gets a lean, responsive setup that adapts to the business’ scale, coupled with tailored support so their team can get the most out of the infrastructure.We push our infrastructure hard and Gcore handles it with ease. The combination of raw GPU power, availability, fast and easy provisioning, and flexible scaling lets us move as fast as we need to. What really sets Gcore apart though, is the hands-on, personalized support. Their team really understands our setup and helps us to optimize it to our specific needs.James Stewart, DevOps Engineering Manager, Leonardo.AiDelivering real-time creation with top-tier AI infrastructurePartnering with Gcore helps Leonardo to maintain its famously rapid pace of development and consistently deliver innovative new features to Leonardo.Ai users.With Gcore, we can spin up GPU nodes instantly and trust that they’ll work reliably and consistently. Knowing that Gcore has the capacity that we need, when we need it, allows us to quickly and confidently develop new, cutting-edge features for Leonardo customers without worrying whether or not we’ll have the GPUs available to power them.James Stewart, DevOps Engineering Manager, Leonardo.AiThe team now uses Terraform to provision GPUs on demand, and containerised workflows to “go fast” when deploying the suite of Gcore AI services powering Leonardo.Ai.Powering global AI creativityGcore GPU Cloud has become part of the backbone of Leonardo AI’s infrastructure. By offloading infrastructure complexity to Gcore, the Leonardo AI team can stay focused on their customers and innovation.Our partnership with Gcore gives us the flexibility and performance to innovate without limits. We can scale our AI workloads globally and keep our customers creating.James Stewart, DevOps Engineering Manager, Leonardo.AiReady to scale your AI workloads globally? Discover how Gcore’s AI services can power your next-generation applications. Find out more about GPU Cloud and Everywhere Inference, see how easy it is to deploy with just three clicks, or get in touch with our AI team for a personalized consultation.
Riga Technical University accelerates genomic research with Gcore GPU Cloud
We saw a 95% reduction in processing time, but more than speed, we also gained flexibility. We could scale from 2 to 8 GPUs instantly, and because usage was on-demand, we only paid for what we needed.Andris Locāns, Head of RTU HPCCompany backgroundRiga Technical University High-Performance Computing Center (RTU HPC) is Latvia’s largest supercomputing resource provider, supporting scientific and technological advancements across the Baltic region. RTU HPC has collaborated with multiple research institutions, including the Latvian Biomedical Research and Study Centre (BMC), a leader in molecular biology and biomedical research. BMC’s genomic research focuses on analyzing thousands of human genomes as part of European initiatives.Accelerating AI-powered genomic processing without compromising controlGenomic research is essential for understanding human health and disease origins, but like any activity that requires sizeable data-set processing, its computational demands are immense. In Latvia, the Riga Technical University High Performance Computing Center (RTU HPC) is leading a shift from traditional scientific computing toward an AI-first model of innovation.Working alongside the Latvian Biomedical Research and Study Centre (BMC), the team set out to solve a critical challenge: rapidly process thousands of human genomes using AI, without losing time or control to hardware bottlenecks or foreign cloud vendors.Essentially, what we wanted was to accelerate variant calling, the computational process of identifying genetic variations.Edgars Liepa, Scientific Assistant, BMCTraditional CPU-based computing often struggles with large-scale genome sequencing and analysis, leading to extended processing times. As a result, the RTU HPC faced several key challenges:The need for faster genome sequencing to support biomedical research.High compute requirements for analyzing large datasets efficiently.The difficulty of sourcing high-performance GPU hardware within a short timeframe.Ensuring cost-effective and scalable computing solutions without major upfront investments.GPU-as-a-service (GPUaaS) for genomic researchRTU HPC turned to Gcore, provisioning Cloud GPUs for immediate access to high-performance computing. Instead of waiting months for on-premises GPU hardware, they gained on-demand access to NVIDIA’s most advanced GPUs—including the H100, designed for AI inference at scale, and located on Gcore’s European cloud infrastructure.“We didn’t want to offload sensitive health data to platforms outside our legal jurisdiction,” Andris Locāns, Head of RTU HPC explains. “Gcore extensive infrastructure enabled us to maintain data sovereignty and compliance by keeping our data in-region, while still delivering the AI acceleration we needed.”This immediately unlocked the following benefits for the team:Instant access to powerful GPUs: Avoiding long procurement cycles for physical infrastructure.Scalability & cost-efficiency: Gcore’s pay-as-you-go model allowed RTU HPC to allocate resources flexibly based on research demands.Data sovereignty: Ensuring genomic data remains within a secure, compliant cloud infrastructure in the Baltic region.Optimized performance: Benchmarking multiple GPU configurations (V100, A100, L40S, H100) for genomic analysis with NVIDIA Clara Parabricks software.“With Gcore, we had near-instant access to compute that would have taken us six months to deploy internally,” says Edgars Liepa, Scientific Assistant at BMC. “That completely changed the pace of our work.”Benchmarking performance for maximum efficiencyRTU HPC and BMC collaborated with Gcore to conduct extensive performance tests on various GPU configurations. By leveraging Cloud GPUs, they identified optimal setups for accelerating genomic workflows.Comparison of CPU vs. GPU: Genome sequencing that previously took over 650 minutes on CPUs was reduced to under 30 minutes with GPU-powered processing.Testing NVIDIA GPUs: Experiments with GPU configurations provided insights into computational efficiency, with findings indicating that scaling up GPUs did not always equate to faster processing.Future discussions with NVIDIA: Gcore’s collaboration enabled further optimizations in GPU usage for genomic analysis.A comparison of CPU and GPU computing times: while CPU processing time exceeded 650 minutes, it could be significantly reduced to under 30 minutes for all tested configurations when using fq2bam H100Faster, scalable, and cost-effective genomic researchNVIDIA H100 GPUs, provided as-a-service by Gcore, delivered graphics processing units with a compute performance that is revolutionizing cloud infrastructure. They are also specifically designed with the power required for high-performance computing tasks such as computational genomics. “It was important for us to see how fast inference runs on the H100,” says Edgars Liepa, “We didn’t customize the model but instead used one developed by NVIDIA, which was already well suited for our task.”The collaboration between RTU HPC, BMC, and Gcore delivered significant benefits to the research program, including:Significant reduction in processing time: Variant calling tasks were completed up to 50x faster.Cost savings with on-demand GPUs: Eliminating upfront hardware investments while optimizing computing costs.Scalable infrastructure: The ability to dynamically allocate resources based on real-time needs.Data sovereignty and security: Genomic data was processed within a compliant, secure cloud environment.The measured processing times of the tested H100 GPUs in detail: the processing time ranged from 13 minutes for haplotypecaller H100 with 2 GPUs used to just under 90 minutes for Deepvariant H100 with 8 GPUs used, indicating that GPU overhead can slow down the processing time when the effective memory limit is reached“It’s not just about going faster,” Liepa adds. “It’s about enabling analysis at a national scale. The AI models are there—but without the right compute power, they’re just theory.”Advancing high-performance genomics in the Baltics and beyondBy leveraging Gcore’s Cloud GPUs, RTU HPC has established a model for scalable, cost-effective genomic research, and this is only the beginning. Now that the speed and flexibility has been proven on genomics processing, RTU HPC plan to broaden AI applications even further.Wider adoption of Cloud GPUs in genomics: RTU HPC is considering Cloud GPUs expansion for broader research applications.Future collaboration with Gcore: Optimization of GPU configurations will continue, and RTU HPC plans to explore AI Inference opportunities with Gcore Everywhere Inference for genomic workloads.Global implications: BMC’s work with the 1+ Million Genomes Project, an EU-wide initiative to make genomic information more accessible for diagnosis and treatment, contributes to international research efforts.The benchmark test results for H100 GPUs using 2, 4, or 8 GPUs: fq2bam and haplotypecaller achieved the shortest processing times on average and comparatively consistent results across 2, 4, and 8 GPUsPioneering AI-powered genomics with sovereign cloud infrastructure“This is the future of AI in healthcare: fast, flexible, sovereign,” says Liepa. “Gcore gave us the infrastructure to make it real—not just for today, but for what comes next.”As AI continues to transform life sciences, the ability to combine cutting-edge GPU performance, regional data compliance, and on-demand scalability is emerging as the key to competitive advantage—not just for companies, but for countries.“We’re proud to support Latvia’s vision for AI-powered genomics,” says Vsevolod Vayner, Product Director of Edge & AI Cloud at Gcore. “This project is a blueprint for how nations can lead in biotech innovation without giving up digital sovereignty.”Find out more about how Gcore Cloud GPUs can enhance your high-performance computing projects.Try Gcore Cloud GPUs
LetzAI rapidly scales AI-powered image generation for global users
Everywhere Inference reduces the latency of our output and enhances the performance of AI-enabled apps, allowing us to optimize our workflows for more accurate, real-time results.Misch Strotz, CEO and co-founder, LetzAIA futuristic tram journey, blending reality and imagination, created with LetzAI. Destination: To the MoonLaunched as a skunkworks project from within Neon Internet, LetzAI is quickly becoming a go-to platform for high-quality AI-generated images. With a mission to democratize and personalize AI-powered image generation, it has emerged as one of the most popular and high-quality options on the market. To support its rapid growth and scale seamlessly, LetzAI partnered with Gcore for advanced AI and cloud infrastructure.Global GPU shortages threaten to derail a new AI image generation ideaIn 2023, Neon Internet CEO and co-founder Misch Strotz was struck by a clever idea: give Luxembourg residents the power to easily generate local images using AI. Within a month, Luxembourg-focused LetzAI V1 went live.Encouraged by strong local demand, Strotz and his team began working on a global version of the platform. The vision? An opt-in AI platform empowering brands, creators, artists, and individuals to unlock endless creative possibilities by adding their own images, art styles, and products. “Other AI platforms scrape the internet, incorporating people and their content without permission. We wanted to put the choice and power in each person’s hands,” Strotz explains.Before long, the team began working on V2. In addition to generating higher quality and more personalized AI-generated images, V2 would drive consistency across objects, characters, and styles. After uploading their own photos and creating their own models, users can blend them with other models created by the community to create an endless number of unique images.However, LetzAI faced a significant hurdle in training and launching V2—a global GPU shortage. With limited resources to train its models, LetzAI needed a reliable partner to help evolve its AI-driven platform and keep it operating smoothly.Finding a trusted, Europe-based AI partnerIn the search for a fitting partner, Strotz spoke to major vendors including hyperscalers and various Europe-based providers. Meeting Gcore’s product leadership team made the decision clear. “It was amazing to meet executives who were so knowledgeable about technology and took us seriously,” recalls Strotz.Gcore’s approach to data security and sovereignty further solidified the decision. “We needed a trusted partner who shared our commitment to European data protection principles, which we incorporated into the development of our platform” he continues.Maximizing AI efficiency by training on local GPUsLetzAI opted for Gcore’s state-of-the-art NVIDIA H100 GPUs in Luxembourg. “This was the perfect option, allowing us to keep our model training and development local. With Gcore, we can rent GPUs rather than entire servers, making it a far more cost-effective solution by avoiding unnecessary costs like excess storage and idle server capacity,” Strotz explains. This approach provided flexibility, efficiency, and high performance, tailored specifically for AI workloads.LetzAI was able to adapt its app to run in containers, configure model training tasks to run on GPU Cloud, and use Everywhere Inference for image generation and upscaling. “Everywhere Inference reduces the latency of our output and enhances the performance of AI-enabled apps, allowing us to optimize our workflows for more accurate, real-time results,” Strotz says.In just two months, LetzAI V2 launched to serve users around the world. And Strotz and team were already developing its successor.Empowering creativity with scalable, high-performance AI infrastructureWith Gcore’s continued support, LetzAI quickly deployed V3. “The Gcore team was incredibly responsive to our needs, guiding us to the best solution for our evolving requirements. This has given us a powerful and efficient infrastructure that can flex according to demand,” says Strotz.Running V3 on Gcore means LetzAI users experience fast, reliable performance. Artists, individuals, and brands are already putting V3 to use in interesting ways. For example, in response to what LetzAI calls its ‘AI Challenges’, a Luxembourg restaurant chain prompted residents to create thousands of images using its model of a pizza.In another example, LetzAI teamed with digital agency LOOP to dress PUMA’s virtual influencer and avatar, Laila, in a Moroccan soccer jersey. According to Strotz, “PUMA had struggled to make clothing look realistic on Laila. When they saw our images, they said the result was 1,000 times better than anything they had tried.”That wasn’t the only brand intrigued by V3’s possibilities. After LetzAI posted V3-generated images of models wearing Sloggi underwear, Sloggi’s creative agency STAN Studios asked LetzAI to generate more images for market testing.Always looking for new ways to support creators, LetzAI also launched its Image Upscaler feature, which enhances images and doubles their resolution. “Our creators can now resolve common AI image issues around quality and resolution. Everywhere Inference is pivotal in delivering the power and speed needed for these dynamic image enhancements,” notes Strotz.Platform evolution and AI innovation without limitsAs its models exceed user expectations worldwide, LetzAI can rely on Gcore to handle a high volume of requests. Confident about generating a limitless number of high-quality images on the fly, LetzAI can continue to scale rapidly to become a sustainable, innovation-driven business.“As we further evolve—such as by adding video features to our platform—our partnership with Gcore will be central to LetzAI’s continued success,” Strotz concludes.
Subscribe to our newsletter
Get the latest industry trends, exclusive insights, and Gcore updates delivered straight to your inbox.