Customer story · ZYNG AI
How ZYNG AI edits 5 million images a month on Jarvis Labs
One GPU cloud for model training, always-on inference, and automated capacity that follows demand.
5M+
images edited every month
3
production compute modes
~10%
higher throughput reported
The workload
One platform, three compute modes.
ZYNG AI helps marketplaces and e-commerce platforms turn raw product photos into listing-ready images. Its pipeline needs different infrastructure at different moments, without splitting the operation across providers.
- 01
Model training and fine-tuning
H100 nodes for custom and open-source models.
ZYNG AI trains small models of its own and fine-tunes open-source models on NVIDIA H100 nodes.
Explore H100 GPUs - 02
Committed inference
Long-running RTX PRO 6000 Blackwell capacity.
Several NVIDIA RTX PRO 6000 Blackwell GPUs run production inference under long-term commitments.
Explore RTX PRO 6000 - 03
Demand-based autoscaling
On-demand and spot GPUs controlled through the CLI.
Automated CLI and SSH workflows add or remove containers and VMs as traffic changes, then deploy and track each GPU. ZYNG AI reports reliable CLI uptime and fast launches for both environments.
Explore the CLI
Why Jarvis Labs
Tested where it matters.
Before standardizing on Jarvis Labs, ZYNG AI ran control experiments across providers. The team compared container and VM launch times, upload and download speeds, CPU performance, memory, and storage.
“Jarvis Labs has been one of the best cloud providers we tested, increasing our throughput by roughly 10% and beating almost all neoclouds we compared.”
— Rishabh · Co-founder, ZYNG AI · Primary compute provider for 6 months
Results and comparisons are reported by ZYNG AI and reflect its workloads and testing environment.
Build on Jarvis Labs
One cloud for every stage of your workload.
Train on high-end GPUs, reserve production capacity, and automate the rest through the CLI.