How NeuroPixel.AI Is Redefining The Fashion Industry With Its GenAI Growth Levers

India’s $110 Bn fashion industry has thrived on innovation, continually setting new trends and pushing boundaries. Recent innovations in the fashion industry, particularly driven by ecommerce platforms, have significantly reshaped how we pick and choose apparel or other lifestyle products and accessories.

Even though the industry has been experimenting with various technologies such as basic AI, metaverse, and augmented reality (AR) for a long time now, it has yet to fully explore the potential of generative AI (GenAI). 

Imperative to mention that, with the evolution of GenAI, brands can now automate the cataloguing process, generate synthetic models, automate and improve virtual try-on experiences, and offer personalised shopping experiences. 

However, concerns linger among brands about unrealistic AI-generated images. Deeptech startup NeuroPixel.AI bridges this gap with its proprietary GenAI algorithms. Besides, the startup also helps personalise the fashion experience for ecommerce businesses.

Founded in 2020 by Arvind Nair and Amritendu Mukherjee, NeuroPixel.AI offers AI-enabled fashion cataloguing, synthetic model generation, and virtual try-ons. 

To do this, it deploys advanced AI/ML and statistical learning theory in computer vision and image processing areas for online retail storefronts. 

The Bengaluru-based startup competes with the likes of OSlash, Vue.ai, Chargebee, and SaaS Labs in the broader Indian deeptech SaaS space.

 

As per Inc42’s “Indian Tech Startup Funding Report 2023,” deeptech startups raised $496 Mn in 2023, up from $397 Mn in 2022. Overall, these startups secured over $1.5 Bn in 343+ deals between 2014 and 2023. 

NeuroPixel.AI has raised approximately $1.2 Mn since its inception. In 2022, it secured INR 2.3 Cr ($299K) in a Pre-Series A round from Flipkart.  

The startup is backed by names like Inflection Point Ventures, Entrepreneur First, Huddle, DLabs, Dexter Angels, etc.

What’s In The NeuroPixel.AI Tech Stack

Today, the use of synthetic models by brands has increased exponentially. While global brands like Nike, Calvin Klein, and Samsung leveraged this trend for several years, India joined the bandwagon only in 2022, with its first virtual influencer, ‘Kyra’, followed by Myntra’s virtual influencer ‘Maya’, and then Naina Avtr developed by Avtr Meta Labs (AML).

For the uninitiated, synthetic models, also known as digital or virtual models, are computer-generated characters that can be customised to resemble anything, possessing the same level of detail as real models.

The GenAI startup, NeuroPixel.AI, uses synthetic model generation technology to improve customer experiences. With this tech, it helps businesses create precise 3D models of their products. This allows customers to see how products look and fit before making a purchase. This increases their confidence in the product, leading to higher conversion rates.  

Nair told Inc42 that while several apps can generate synthetic humans with background alterations while keeping clothing constant, the difficulty lies in maintaining consistency within generative AI solutions. 

“Existing solutions like MidJourney, Standard Stable Diffusion Models, and Gemini offer beautiful outputs, yet struggle to uphold consistency when building around a specific element,” he added.

Despite the emergence of startups like Modelverse, GetAyna and AlphaBake in this space, Nair believes that the quality offered by the startup surpasses that of its competitors. 

“Our focus on advancements in synthetic human generation such as the quality of skin texture, eyes, fingers, toes and lightmaps is a major giveaway for stable diffusion-like models,” Nair said.

Moreover, as part of their process, every virtual influencer created by the startup undergoes facial recognition checks using a third-party tool to ensure it doesn’t resemble any existing face online. 

Besides, the startup uses open-source models in its technology stack but has its own IP across multiple layers.

“Building everything from scratch isn’t practical, so we leverage various open-source APIs. For instance, we utilise standard stable diffusion models for background generation and state-of-the-art segmentation algorithms developed by Meta. However, approximately 50% of our technology stack comprises our core intellectual property, including a few patents,” Nair added.

He mentioned that the startup has filed patents related to synthetic human generation, attribute modification, and apparel warping.

For synthetic humans and similar aspects, the brand utilises pre-trained open-source models that are generated based on its training data. 

“Instead of directly accessing training data, we rely on pre-existing models. Our approach involves a combination of our proprietary data from our own shoots and external data from open-source algorithms.”

How Does NeuroPixel Make Money?

Currently, the startup is also leveraging GenAI to automate the apparel cataloguing process. With this, it helps businesses cull the time spent on clicking photos manually and editing images, making the process of adding new products to their online stores seamless.

According to the startup, automated cataloguing helps small fashion businesses reduce errors, enhance efficiency, and improve customer experience, all while competing with larger players.  

“Outdoor shoots can be costly for brands due to the minimum order quantity requirements and logistical challenges. Our approach eliminates these constraints. Flexibility is offered in terms of models, styles, and locations, enabling diverse marketing images without the need for a minimum order quantity. This proposition particularly benefits direct-to-consumer brands or platforms that seek cost-effective marketing solutions at scale,” Nair said.

He added that the pricing model operates on a cost-per-image basis, dependent on the volume of images and the required Turn Around Time (TAT). The brand offers two types of shoots: standard catalogue and marketing editorial. 

For standard catalogue shoots, the cost ranges from INR 600 to 900 per item for four to five catalogue images (front and back). 

“For marketing editorial images, the typical cost per image ranges from INR 4,000-5,000. However, our solution reduces this cost by about 70%, bringing it down to approximately INR 800 to 1,000 per image,” Nair said.

The startup counts Myntra, Fabindia, Van Heusen, The Pant Project, Decathlon, Yukio, Jolene.club, Soch, and Styli as its clients.

The startup’s business model revolves around charging customers on a pay-per-image basis, with discounts offered for larger volumes. However, within the next three months, it plans to transition to a subscription model.

“Under this new model, users will subscribe and receive a set number of credits, such as 200, for a fee, typically around $100. Each image download will deduct a credit from the user’s balance. Additionally, users will have the flexibility to top up their credits during the month, particularly if they are heavy users. This subscription model offers a more flexible and cost-effective approach for our customers,” Nair said. 

Enough Scope Ahead?

From synthetic model generation to automated cataloguing processes, the startup is trying to reshape the way ecommerce businesses engage with customers online. However, challenges persist, particularly in accessing crucial GPU infrastructure.

As the industry continues to bloom, the biggest challenge for NeuroPixel.AI is access to graphics processing units (GPUs) for training algorithms. 

Illustrating this, Nair adds that the two recent entrants into virtual try-ons are Alibaba Cloud and Google and both have access to massive GPUs and cloud infrastructure. 

“In just one training run, they use 400 times more computing power than we have had access to in our three years as a startup. Their ability to train large models far exceeds what’s possible for smaller startups like ours,” the cofounder said. 

According to Nair, to succeed as a startup in generative AI, three things are crucial — great engineering and research talent to develop algorithms, access to datasets, and critical GPU infrastructure. However, securing GPU infrastructure remains a significant obstacle for small AI startups like NeuroPixel.AI which are in the seed stage.

Despite this, NeuroPixel.AI’s potential to grow in the burgeoning Indian GenAI market is unimaginable. 

As per Inc42, the country’s GenAI market is anticipated to surpass $17 Bn by 2030 from $1.1 Bn in 2023, growing at a CAGR of 48%.

[Edited by Shishir Parasher]