Photo credit: venturebeat.com
Whether you’re a fan of Android or iOS, the name PicsArt is recognizable to many. Since its inception over a decade ago, PicsArt has emerged as a leading platform for image and video editing, boasting a user base exceeding 150 million monthly active users.
Nevertheless, the road to success has not been without its hurdles. Although PicsArt was an early pioneer in mobile editing, it now faces fierce competition from established entities like Canva and Adobe. The CTO of PicsArt, Artavazd Mehrabyan, shared his insights on the challenges the company faces during the recent WCIT conference in Armenia. He emphasized that differentiating oneself in a crowded market is a persistent struggle.
“Many features that were once unique to PicsArt have been adopted by our competitors. We were the first to provide an all-in-one editing service on mobile devices back in 2011, but others have quickly followed suit,” Mehrabyan stated. He noted that the landscape is particularly competitive in the domain of AI, where similar functionalities are becoming widespread among various photo editing services.
A case in point is PicsArt’s object generation feature, which utilizes advanced AI to enable users to create specific elements within their images. This feature has since been mirrored in other editing platforms, leading to a notable overlap in offerings.
PicsArt AI GIF generator
Rather than merely expanding its toolkit to stay ahead, PicsArt is concentrating on enhancing the quality of its existing AI capabilities. According to Mehrabyan, the goal is to refine the way these features are presented and customized to meet users’ needs—be it removing an unwanted item from a holiday photo or producing compelling advertisements with both visual elements and copy.
Training high-quality creative AI
In its early days, when AI was not yet a focal point, much of PicsArt’s research centered around creating a seamless mobile editing experience. “Making comprehensive editing features function offline on devices was quite challenging,” Mehrabyan recalled. As the user base surged, the company shifted to a hybrid infrastructure model that combined multi-cloud solutions with data centers, which remains a cost-effective and flexible approach.
With this robust tech foundation, PicsArt launched its first AI feature in 2016, employing smaller models that could operate offline on user devices. This laid the groundwork for a more extensive AI initiative, eventually evolving into an AI-first organization leveraging its backend services to introduce capabilities such as background removal and replacement. More recently, the company has begun to train its own generative AI models from the ground up.
In the creative space, user retention is crucial. A minor misstep that results in subpar quality could lead to losing a user for good. To mitigate this risk, PicsArt places a significant emphasis on the data behind its AI. The company harnesses data from its user network, specifically those marked as public and available for editing.
“Our special ‘free to edit’ license allows users to tag their images as available for reuse, contributing to a communal pool of content,” Mehrabyan explained. This long-standing license has enabled PicsArt to amass a wealth of user-generated material for training its AI, although not all of it meets high standards. Consequently, the data undergoes rigorous cleansing and processing to become a viable training resource.
“As a result, we possess a substantial proprietary dataset that does not necessitate external data acquisition,” he added.
However, simply having high-quality data is only part of the equation. The real challenge, as described by Mehrabyan, was establishing what he refers to as the “data flywheel.” This self-sustaining cycle not only addresses data accessibility but also factors in annotation, usage, and an ongoing learning process to enhance AI capabilities continuously.
“We developed our own annotation technology to create an efficient infrastructure for image classification, tagging, and labeling,” Mehrabyan stated. “A dedicated team refines this pipeline, combining automatic processing with human input for continuous improvement.”
Feedback loop leads to contextual intelligence
The combination of human feedback and AI-driven processes has been essential for enhancing the quality of outputs that PicsArt generates. This focus is steering the company toward what Mehrabyan terms “contextual intelligence” — the ability of the platform to discern user needs and deliver precisely what they seek.
This capability holds significant value for the platform’s expanding segment of business users, who are increasingly looking to create content directly from their smartphones. While most users still leverage PicsArt for personal projects, the company has identified a growing demand for work-related applications, especially in marketing.
“Contextual intelligence not only tracks user activity but also predicts their next steps, functioning both reactively and proactively,” he elaborated. This means that users creating work-related content won’t need to establish brand guidelines each time, as the platform will already have the necessary context to generate fitting outputs. Plans for a brand kit feature will allow users to customize this context further, enhancing content quality.
Creative AI agents on the way
Mehrabyan envisions that advancements in contextual intelligence will pave the way for an agent-based ecosystem at PicsArt. This would provide users with an intelligent assistant that is well-versed in their design preferences and historical context.
“This assistant would grasp user intent and context, offering interactive guidance to enhance productivity,” he shared. Moreover, the integration of AI agents could facilitate bulk task execution, enabling users to automate workflows, thereby streamlining their creative processes.
This forward-thinking approach aims to position PicsArt as a frontrunner in the creative sector, empowering users to enhance their creativity and drive business growth effectively.
Mehrabyan acknowledged that while AI implementation could initially appear daunting, it ultimately has the potential to transform users’ roles. He believes that rather than narrowing their expertise, individuals will leverage AI to broaden their skills and knowledge at a faster pace.
According to projections from Future Markets Insights, the market for AI image editing is expected to rise from $80.3 million in 2024 to $217.9 million by 2034, indicating a CAGR of 10.5%. Additionally, the segment focused on AI-driven generation, which has become integral to many editing tools—including PicsArt—is projected to experience substantial growth, surging from $8.7 billion in 2024 to $60.8 billion by 2030.
Source
venturebeat.com