Can fashion retailers cut costs by relying on AI models?

Can generative AI modelling replace human models?

Last month, Generation did something unexpected. A brand widely popular for its inclusive campaigns, with an intentional focus on representing real bodies and real people, was now experimenting with AI models. 

Online shoppers scrolling through the website were quick to realise that something was different. Before we knew it, X, formerly known as Twitter, was blowing up with users pointing out that Generation seems to be generating AI models of their own. 

The only problem was that AI is not yet sophisticated enough to produce results that would align with Generation’s ethos, which largely combines Pakistani culture and inclusivity. 

When Profit reached out to the brand, they declined our request to comment, saying that it was a small experiment and they would require toying with the AI tool for longer before they can provide insights into the potential of it. 

However, our interest had already piqued, so we sought others who could talk about the potential benefits and drawbacks of replacing human models with AI generated ones.

Artificial intelligence – a double edged sword?

Is Generation the only fashion brand that is dabbling with AI or are there more?

Wasay Hasan, director at Lulusar, told Profit, “Over the past seven to eight years, we’ve closely witnessed technological advancements, with AI emerging as a prominent topic in the last one to two years. At Lulusar, we’re exploring AI’s potential to enhance efficiency, particularly in reducing costs and lead times, a key aspect of our brand identity. We aim to be early adopters, observing various AI applications showcased at events like a fabric procurement fair in Germany.”

He added that, “From pattern making to marketing campaigns, AI offers transformative possibilities, evident in trials where entire collections can be designed in under a month. Our firsthand experience shows the rapid technological strides made in AI within the industry.”

Hasan admitted that Lulusar has not experimented with AI to generate human models, but they definitely plan on doing so soon. “We want to be the first, amongst the first to adopt it in Pakistan,” he said. 

“At Lulusar, we work at a very fast rate. We have a collection come out every week. With AI, we could streamline this process, conducting shoots digitally and rendering them on the same day, enabling immediate publication. This transformative integration could parallel Lulusar’s initial market entry eight years ago,” Hasan divulged. 

He added, “Globally, no fashion brand matches our weekly release rate. AI could enhance our inventory management algorithm, optimising our ability to forecast and clear stock efficiently. Currently, we clear 90-95% of inventory within the launch weekend. AI could further increase efficiency, refining our operations based on data analysis. For instance, during peak periods like Eid, AI could predict increased demand and advise on production volume, potentially surpassing our current output estimations. Through a year-long AI implementation, the system could adapt to our business model, improving order projections and operational efficiency.”

Hasan believes that by aligning AI insights with historical data and business practices, Lulusar could exceed growth projections, potentially achieving 1.5 times our current growth rate annually. He admitted that AI’s potential to revolutionise not only Lulusarr’s operations but also those of other brands by optimising production and forecasting processes.

Commenting on the drawbacks of leveraging AI, Hasan told Profit, “Despite AI’s potential, unethical applications pose concerns due to the lack of regulations, allowing competitors to replicate campaigns. Such misuse can result in misaligned campaigns, targeting the wrong audience or failing to reflect brand identity accurately. Additionally, the misconception of AI’s simplicity undermines its complexity, requiring specific commands and a deep understanding of its capabilities. While AI offers opportunities for innovation, its effective use demands expertise and caution, making it unsuitable for every team. As someone familiar with AI applications like chat GPT, I can attest to the necessity of meticulous commands for desired outcomes, dispelling the notion of effortless implementation. So, while AI holds promise, its ethical and technical challenges necessitate careful consideration before integration into a brand’s operations.”

He also told an anecdote about an applicant who had used AI to replicate Lulusar’s own campaign when putting together his portfolio. “There was an interview where we had someone come and ask us what we thought of their design aesthetic. And maybe you can argue that to their disadvantage, we are a brand that’s very technologically advanced and we know how to use our technology. So, we found out that in this portfolio that was sent to us from one of these interviewees, they just AI generated our whole collection and pasted it onto their portfolio. How were we able to pick on that? I could tell the backgrounds were very similar. The proportions for the windows were the same. The clothes were also the same, just the colours were different.”

Another possible problem is the pervasive accessibility of AI that may pose challenges as it is marketed as a universally applicable solution. Over the next five to ten years, the acceptance and effectiveness of AI applications will become apparent, given its capacity to create visually engaging content that may attract viewers’ attention more for its artificiality than the actual product. 

Hasan, speaking on the matter, said, “A striking ad featuring AI-generated imagery may garner attention but may not necessarily translate into increased sales or engagement. While platforms like Instagram feature novel AI-generated content, its effectiveness in traditional media like billboards or TV ads can not be commented upon yet. Despite AI’s potential to revolutionise industries like fashion, its misuse can perpetuate unrealistic beauty standards, reminiscent of past practices of digitally altering models.” So you might get eyeballs on your advertisement, but not for the right reasons. People might only engage with it because it’s different, not necessarily because the product you are marketing got their attention. 

The fact remains, it is volatile and that raises concerns regarding its influence on societal perceptions and values. If unchecked, AI could reinforce harmful stereotypes and ideals, shaping consumer behaviour in a negative way. It also has the potential to dictate cultural standards, driven by the preferences and biases of its creators. So, careful consideration and regulation are necessary to ensure AI’s potential benefits are maximised while mitigating its negative impacts on society.

Understanding the tool

The most salient feature that helped shoppers identify the use of AI in Generation’s experiment was limbs and fingers. It is, more often than not, details, such as hands and feet that give it away.  If you remember the Kate Middleton fiasco and her unexpected Mother’s Day post that sent the internet into a frenzy, you would know that getting the smaller details of the human body is something that artificial intelligence is not so intelligent in replicating. 

Profit asked Soban Raza, CEO of AI company Antematter, why AI always gets the hands wrong? He explained, “It’s widely acknowledged that in AI datasets, human hands are often less prominently featured compared to other body parts. They tend to appear smaller and less frequently in source images, making them challenging for AI models to grasp. While AI learns and replicates patterns, it lacks a true understanding of concepts like finger count, focusing instead on pattern recognition and arrangement.”

He continued, explaining that as a consequence of AI’s lack of understanding of the aforementioned characteristics, if hands or finer details are poorly represented in the dataset, it’s likely that current AI models struggle to generate accurate images. “The variability in human body movements, influenced by factors like age, gender, and individual characteristics, further complicates matters. Additionally, in motion pictures, body parts interact intricately rather than moving independently, posing another challenge for AI comprehension. Subtle nuances, such as the impact of a slightly bent finger, often go unnoticed in datasets, hindering AI’s ability to learn such patterns effectively. Consequently, hands, fingers, and subtle movements tend to be overlooked by AI models,” he concluded. 

But is the flaw inherent to the model itself, or is it due to the limitations of the commands provided?

We asked Raza if sophisticating our commands can solve this issue. If we were to feed more detailed instructions and provide distinct components separately, such as feeding pictures of hands separately from images of the full human figure, will the dataset become clearer?

Raza answered, saying, “This question is a focal point for many well-funded startups and agencies, attempting to strike a balance between dataset refinement and model optimization. Correcting training data might help AI models to get the finer details right, however, such commands may risk model overfitting. This is where proficiency in one task compromises others.”

We learnt that aggressively refining training data may lead to overfitting, focusing on hands and fingers at the expense of other body parts or generative capabilities. 

Current AI tools are not advanced enough to perform both these tasks, however, companies dedicated to building and improving AI tools are working to strike the perfect balance. 

Another issue you are likely to run into when using AI to generate images, especially of humans and certain cultures, is that AI is largely unregulated and frankly racist. It absorbs internet data leading to stereotypes, often generating images that are inaccurate and biassed, for example traditional Indian women in orientalist scenes. Can this problem be tackled by teaching AI broader, culturally nuanced perspectives beyond traditional, biassed portrayals? 

While responding to this question, Raza highlighted, “I think what you mentioned is a very challenging problem. It’s not a problem that is for one startup to solve. It’s actually at the core of the biggest challenges that are currently attacking AI, so to speak, because think about it. The problem that is generative AI has been built from this massive amount of data– the Internet.”

So if we complain that the data fed into AI is biassed, correcting this bias at such a massive scale is insanely complex. Raza elucidated, “So just to make the consumption of this data on the internet possible for creating image, video and audio files took us decades, right? Consumption of such a scale of data to build something was actually remarkably difficult. Now, to debias it and remove stereotypes at that scale presents an open challenge.”

According to Raza, this is an open technological problem that is being worked on at this point in time. Huge loads of money is being thrown at it. “There is no easy answer to say that we should do this or that. It’s just going to unfold as regulations unfold, as technological advancements unfold, maybe big companies like open AI can figure something out on that front. But I think you’ll have to keep worrying about it for the time being.” 

When asked what he envisions as the trajectory for developing nations, like Pakistan, to harness AI effectively and how can we transition from conventional methods to AI-driven solutions, such as image generation and other generative services, Raza said, “Businesses can gain a significant competitive edge by effectively integrating AI into their workflows, particularly in industries like IT services. By enhancing internal processes with AI, companies can boost productivity and outperform competitors, securing business opportunities they might have otherwise lost.”

He informed that currently, AI serves primarily as an assistive technology, streamlining tasks and increasing operational efficiency by up to 30-40%. The focus lies on improving internal operations rather than developing AI products for external markets, which still requires significant development. Initial steps involve leveraging AI to enhance productivity and operational excellence before expanding into AI product development. Regulatory challenges and other external factors are currently less relevant in the Pakistani economy compared to internal productivity improvements through AI integration.

So, businesses should prioritise optimising internal processes with AI before exploring broader market applications.

We also asked Raza whether fashion brands are a target market for companies like Antematter based in Pakistan, to which he replied, “Many businesses rely on AI software for tasks like trend analysis, market patterns, or accounting, catering to a broad range of industries. However, when it comes to human image generation for ad campaigns, particularly in the fashion sector, the focus may not be as prominent. This isn’t due to a lack of interest from companies like ours but rather a need to understand how such ventures would unfold.”

He confirmed that if fashion brands in Pakistan show interest and traction in this area, AI companies would undoubtedly shift focus. Raza also pointed out a potential issue with this, “The complexity lies in addressing challenges like high-quality training data, transforming 3D data into 2D, and capturing movements effectively. The sophistication of technology, real-time processing capabilities, and attention to model fidelity are critical considerations. The readiness of Pakistani fashion brands to embrace AI will ultimately dictate the pace of progress in this direction, given the country’s historical cautiousness in adopting cutting-edge trends.”

After all these considerations, we concluded that AI, while bursting with the potential to revolutionise content creation, curation and marketing, it will take quite a lot of investment, in terms of capital, efficiency and practice to be able to smartly use the tool for generating AI models and images.

Nisma Riaz
Nisma Riaz
Nisma Riaz is a business journalist at Profit. She covers tech, retail and marketing and can be reached at [email protected] or


Please enter your comment!
Please enter your name here

Popular Posts