Wired editorial staff were not involved with the creation of this content.
Artificial intelligence (AI) has shifted how industries operate, and fitness is no exception. By bringing lab-grade motion analysis into everyday gym settings, real-time biomechanical AI is transforming traditional strength training. With emerging solutions like Flex AI’s form-checking technology serving as a case study, serious lifters could leverage the analytic potential of AI to achieve results.
Going through a workout routine might seem simple, but variations in form could result in poor results at best and injury at worst. Serious lifters might use a recording device to check if they are using the proper form, but this only helps users make adjustments after the fact. It doesn’t help if a lifter feels bad under the bar. When pain appears, it’s possible that they’ve been following an improper pattern for months.
Existing form-checking apps aren’t necessarily analytical, Flex AI founders Amin Niri and Amol Gharat suggest, but a guess based on averages. These applications may be able to demonstrate the perfect form, but every lifter is built differently. To perform actual biomechanical analysis and catch specific errors in real time, a new kind of personal training solution is necessary.
Basic pose estimation might be helpful, but it isn’t usually the kind of information that serious lifters are looking for. Actual infrastructure built on robust data is essential for progress in this area; however, innovators like Flex AI have found that existing datasets are somewhat lacking. Medical labs have motion capture data and expensive systems unsuitable for real gym environments, and video streaming platforms have millions of lifting videos that aren’t properly annotated with the detail an AI needs to learn proper form.
Despite this challenge, the need for real-time feedback systems to transition from lab environments to gym settings has become apparent. After all, analysis of proper form is only beneficial if examples of that form may be properly practiced. Companies like Flex AI recognized the gap between less accessible medical analysis and freely available video examples, and resolved to join these with AI.
To create the dataset necessary for an effective form-checking tool, Flex AI started from scratch, analysing thousands of hours of exercise footage, tagging each frame for dozens of biomechanical markers such as hip angle, knee tracking, spinal position, bar path, weight distribution—everything that matters when a lifter is moving a heavy weight.
Capable of accounting for lighting, camera angle, passersby, and even equipment in the way, Flex AI tracks points per frame to process form directly on the user’s phone.
“We had to optimize every layer of the system,” Flex AI’s head of AI Gharat explained. “Standard computer vision frameworks weren’t built for this. We needed real-time feedback in uncontrolled environments, running on phones, with zero latency. That took years of engineering.”
The real-time motion tracking and biomechanical modeling capabilities of AI systems might be promising, but the question of efficacy remains. There are already a number of form-checking apps, even examples powered by AI, that do not necessarily reflect an improvement in form. Compared with exercise videos, however, experts suggest that leading solutions could have improved results.
A recent study explored whether deep neural networks could support a mobile-based personal workout assistant. The participants of the study had no prior squat experience and were divided into groups, one with a mobile workout assistant and one without. The experimental group that used mobile devices reported improved squat postures.
.png)
Before launching its product, Flex AI underwent extensive testing with experienced lifters. These were individuals who had been training seriously for five or more years, and Flex AI founder Niri claims that even they found the results to be humbling. Many were discovering form issues that they didn’t know they had, such as hip shifts at the bottom of squats or spinal rounding on deadlifts.
“I thought my squat form was dialed in,” one beta tester reported. “Turns out I’ve been shifting my weight onto my left side at the bottom for who knows how long. Never felt it. The AI caught it immediately.”
In a similar vein to the study above, Flex AI claims that AI-supported beta testers reported measurable improvements within just two weeks of use. Results included better depth, cleaner bar paths, stronger lifts, and reduced risk across major movements. For lifters who prefer to train alone or can’t afford coaching, Flex AI promises change.
AI certainly appears to be more effective than existing examples, but it is unlikely that the technology would outperform a personal trainer. Regardless, the accessibility of AI compared to personal training solutions may have broad appeal among both established and aspiring lifters. As a new technology, the limits of AI-based form correction have yet to be explored, but its potential is apparent.
Form-checking solutions are only a single part of strength training processes, and AI may play a greater role in this space. Adaptive feedback, predictive modeling, and wearable AI solutions could each contribute toward a more holistic system of strength training. As leaders like Flex AI establish their solutions in the market, they may bolster related developments and research in the field.
