AI and Fashion: It Does Not Replace the Craft, It Puts It to the Test
On screen, the trousers worked. I realised they needed correcting precisely when there was no longer an obvious mistake to find.
I was designing a pair of women’s trousers in technical fabric, with a harem silhouette, quilting and zips along the sides. I used Marvelous to observe how the material would behave and identify potential fit issues.
Marvelous is a 3D physics-based simulation tool. It processes the pattern, avatar and assigned fabric properties. It can form part of an AI-assisted workflow, but the reliability of the result depends on the accuracy of the data entered.
I had described the fabric, construction and finishing in detail. The simulation produced a result that was consistent with the material and close to the concept. The volume was clear, the design details sat correctly within the structure and the trousers appeared resolved.
As I continued to study the fit, the crotch line began to look unbalanced in relation to the rest of the garment. I reduced its width while preserving enough room for an unrestricted stride. I also adjusted the side line.
I wanted the upper section to retain the clean, controlled construction of a classic pair of trousers, while the shape below the hips became softer and more deconstructed. Left unchanged, the volume through the crotch would have made the garment look heavy and shifted the proportion.
The software had processed the information correctly. I was reading something else: how that construction would behave on the body, the relationship between volume and movement, and the possibility that a fit which looked interesting on screen might become unbalanced once worn.
I had encountered similar problems during fittings and knew that reducing the crotch without checking freedom of movement could make the trousers feel restrictive. The correction had to remove volume in the right place without erasing the character of the design.
Marvelous helped me reach that decision sooner. I could assess proportions and design details without waiting for a physical sample, then intervene while the changes were still manageable. The concept was not adapted to the software. I used the software to understand more clearly where I needed to act.
For me, this is the most useful role AI can play in fashion: shortening parts of the process and making more information available before committing time, materials and production resources.
Across the industry, the most tangible benefits often appear in less visible areas. AI can combine sales, availability, seasonality and returns data to improve forecasting and product allocation. H&M reports that AI-driven forecasting and RFID technology have made delivery planning more accurate. Zalando states that in 2025 its Size & Fit systems prevented around 8% of size-related returns.
These results affect costs, inventory and margins. They also reveal where AI performs best: large quantities of data, repeated operations and patterns that are already documented well enough to be compared.
In creative research, the outcome is harder to measure. I use AI to analyse trends and benchmarks, connect information and test a direction. In a few hours, I can examine scenarios that would previously have taken days. The time saved is real. So is the risk of confusing an abundance of proposals with the quality of the thinking behind them.
Generated images arrive already styled, coordinated and visually convincing. They give a concept a sense of completion that the product has not yet earned. None of those images has to meet a fabric consumption target, maintain its proportion across different sizes, withstand washing or prove that the construction will hold.
Virtual try-on, however advanced it becomes, is still a representation. It can recommend a size and help reduce some returns, but it cannot assess the pressure of a waistband, the mobility of an armhole or the feel of a fabric against the skin. A customer may see an image of themselves wearing a jacket without knowing how that jacket will behave when they raise an arm.
Trend analysis requires the same caution. An algorithm can identify growing signals and recurring behaviours. Its raw material is what has already happened. When different brands use the same tools, start from similar references and follow the same indications, their proposals quickly begin to converge. The result may be correct and current, while looking remarkably similar to a competitor’s.
The issue also affects professional development. Researching, comparing, checking and correcting are part of how people learn the craft. Delegating these activities entirely may produce a first answer more quickly, without ensuring that the person receiving it knows how to recognise an error. Speed cannot compensate for judgement that has never had the chance to develop.
There are other, less visible concerns: protection of designs, confidentiality of uploaded materials, the origin of images used for training and the energy consumption of AI systems. Before introducing AI into company processes, businesses need to know what data is leaving the organisation, where it is being stored and who retains control over the output.
AI cannot feel the hand of a fabric, enter a fitting room, understand a supplier’s actual capabilities or take responsibility for a price, inadequate quality or a collection that has lost its identity. It can reduce uncertainty, organise data and bring us closer to a decision more quickly.
In the trousers I was developing, the software had already done its part well. The decisive change involved just a few centimetres along the crotch and side lines. I corrected those. Everything else could remain exactly as it was.
Romina Tosi
Disclaimer
The views expressed above represent my personal interpretation of publicly available information and, like any interpretation, may be shared, debated, or challenged.
The information referenced comes from public sources available at the time of publication, including official documents, union communications, press articles, and materials accessible to anyone. I do not disclose confidential information or facts learned through privileged access. I do not reveal protected information, nor do I attribute unlawful conduct to individuals or companies.
Any reference to specific cases is intended solely to provide context and analyze dynamics affecting the broader industry. It is not intended to target individuals or particular businesses.
Observations regarding industrial strategies, financial decisions, and production models fall within the right to express opinions and commentary on matters of public interest. They remain personal assessments, not definitive judgments.
Not all companies operate in the same way. Alongside businesses that may deserve criticism, there are many others that work with seriousness, consistency, and long-term vision.
If you notice any errors or inaccuracies, please let me know. I will be happy to review and correct them where necessary.
The purpose of this reflection is to encourage discussion and debate, not to cause harm to individuals, companies, or organizations.
Comments