How Gepetto Achieved Faster Cold Starts While Cutting Infrastructure Costs

$1,000s
Saved Monthly
5x Faster Cold Boots
Compared to other providers
>99.9%
Production Uptime

“Switching to Beam has been really helpful. It has addressed all the pain points we had with Replicate, enabling us to provide a better, more reliable service to our customers.”

Simon Brami
Founder at Gepetto

About Gepetto

Gepetto is an AI-powered virtual staging and interior design app that helps people visualize and create their dream spaces. Using Gepetto, a user can upload a photo of a room, select an architectural style or design for the space, and generate a realistic rendering of their room in the chosen style.

Gepetto aims to help real estate agents and property owners enhance their listings, making spaces more appealing to potential buyers. The platform also includes features for customizing specific areas and improving outdoor spaces.

Inspiration Behind Gepetto

Simon Brami was inspired to start Gepetto out of a passion for interior design and a fascination with AI. His goal was to create an app that would help users bring their design ideas to life through realistic visualizations.

Migrating From Replicate

When Simon started Gepetto, he used Replicate for AI inference. Simon’s initial attraction to Replicate was its straightforward pricing model, which billed only for the prediction time. However, earlier this year, Replicate suddenly changed its pricing model, leading to a dramatic increase in Simon’s operational costs for Gepetto.

"They sent me an email that went into my spam folder and told me that the way I was using Replicate was not explicitly against their terms and conditions but the way I was using Replicate wasn’t profitable for them, because I was pinging the model to keep it idle and avoid long boot times."

In the email, Replicate announced that their new billing policy would charge users for time their private model was idle, rather than just the model prediction time.

“I suddenly found myself paying for every second the server was on, even if it wasn’t making predictions,” said Simon. “My bill went through the roof.”

Beyond billing, Simon faced several other challenges with Replicate:

• Slow Boot Times: Replicate's GPU machines had long boot times, especially for the heavy pipelines Gepetto must load, often taking up to five minutes to load the models. This delay was unacceptable for Gepetto’s customers, who demanded faster results from their image uploads.
• Security Concerns: Using public models on Replicate posed security risks. Public models allowed anyone to pull the Docker image, exposing Simon's application code for others to use.
• Downtime and Service Outages: Simon experienced multiple issues with downtime on Replicate. Frequent downtime and machine booting problems directly affected Gepetto’s customers.

Simon needed a solution that would offer faster boot times, more reliability, and better overall performance.

Achieving Better Performance and Pricing with Beam

Simon found several compelling reasons to switch to Beam:

• Faster Boot Times: One of Simon’s main frustrations with Replicate were the long boot times. "Replicate was taking a minimum of two minutes to spin up the machine. With Beam, even though my models are pretty huge, they still load a lot faster than they did on Replicate.”
• Better Reliability: Reliability was a major factor in Simon’s decision to switch hosting providers. "Recently, Replicate has had a lot of issues with downtime. With Beam, my apps are running perfectly and the resiliency has been much higher,” said Simon.
• No Charges for Idle Time: Beam’s billing model also helped Simon, because he no longer pays for idle time, as he did on Replicate. "In the end, I have a pretty normal bill at the end of the month, which is the equivalent of my previous bill in Replicate. But with Beam, the inference is so much faster including the boot time."
• Ease of Integration: Transitioning from Replicate to Beam was easy. "The onboarding was exceptional—phenomenal. I've never seen an onboarding that easy for a developer. Within minutes, I already had an app deployed."

Since Simon migrated to Beam, Gepetto’s users are getting better performance and a more reliable service.