To resolve these architectural hurdles, Picsart chose to build its core environment on Google Cloud nearly nine years ago, establishing a trusted, single-vendor relationship. To store the massive daily influx of user media safely, the team deployed Google Cloud Storage, while running their virtual infrastructure on Google VMs.

Adhering to the principle that technology should remain completely invisible to the user, Picsart crafted sophisticated AI chains. When a user enters a basic, short description, a backend Gemini reasoning model acts as an agent to automatically rephrase and enhance the prompt based on optimized prompting techniques discovered in Google’s technical documentation.

If a user wants a cartoon animation of themselves, they simply upload a selfie. Google Nano Banana instantly transforms that photo into a cartoon style, establishing the animation's foundational first frame. This frame is then combined with the Gemini-enhanced prompt and fed directly into the Veo video-generation model to output seamless, high-quality video clips. To take this further, the team built "Picsart Flow"—a highly visual tool where creators connect multiple modular nodes, such as image generators, video processors, dubbing tools, and music creators, into a unified, automated asset pipeline.

"We used a Gemini reasoning model that would take the user prompt, which is very short, and then rephrase it in a way that's the best for the Google model,” says Narek Hayrapetyan, Head of Product at Picsart. “It's happening in the back end without hindering user experiences."

Picsart has also begun testing the Gemini Omni, a new multimodal video creation and editing tool, behind the scenes with a particular focus on prompt-based editing.

“This is a fundamental game-changer for how creators communicate their visions to Picsart,” says Hayrapetyan. “With a simple text prompt, Gemini Omni allows users to execute complex VFX, like swapping out a character or transforming the environment, while flawlessly preserving the core elements of their footage.”

Picsart has found that Gemini Omni has been especially useful in post production thanks to the ability to adapt content on the fly, as well as text control given its ability to generate and edit text within videos–a capability generative models had up until now struggled to achieve.

“Multimodal generative AI is undeniably the future of video creation,” adds Hayrepetyan. “With Gemini Omni, we’re thrilled to bring this powerful, granular technology to 130 million Picsart users worldwide.”

By anchoring its operations to Google Cloud's cohesive stack that includes Google Cloud Storage, Google Kubernetes Engine (GKE), and Agent Platform, Picsart has seamlessly integrated over 120 unique AI models into its platform. The engineering team collects billions of data events every day to execute rapid, continuous A/B testing at an immense scale. These exhaustive tests provided a surprising consumer insight: users consistently prefer ultra-fast models over slower, marginally higher-quality options.