What is happening behind the 90% reduction in photography costs? AI models shaking the fashion industry

A garment found on an online shopping site. After observing the silhouette when worn by the model and checking the shoulder drop and length, I press the purchase button.

However, what if the person in the image wasn't actually wearing the clothes? Furthermore, what if the person in the image didn't exist at all? What clues should we rely on to choose clothes?

The introduction of AI in the fashion industry is advancing beyond just enhancing the appearance of photos, transforming human faces and bodies into reusable data.

A Reuters article published by iTnews on October 27, 2026, followed this change from both contractual and employment perspectives. What emerges is the allure of increased efficiency in photography and the underlying issue of "who controls one's image."

AI models present two distinct issues

Even when we talk about AI models, their reality is not uniform.

One method involves generating fictional models not based on specific real individuals. Brands create images of people that match the product's vibe instead of hiring human models.

The other method involves creating digital doubles of real models based on their photos or appearances, or editing pre-existing images. This includes overlaying faces on different bodies, changing expressions, or making it look like they're wearing different clothes.

In the former, the main contention is the replacement of jobs previously commissioned to humans. In the latter, additional issues of consent, extra compensation, and the scope of appearance usage arise.

Even if a likeness appears in an advertisement, it doesn't necessarily mean the person participated in a new shoot. Opportunities for negotiation and compensation that traditionally arose with each shoot might be absorbed into the initial contract.

Contract clauses on "face swapping" and "no additional compensation"

Reuters interviewed 15 heads of model agencies and examined over a dozen confidential shooting contracts.

Among them were clauses requesting face swapping using AI, creating digital replicas without additional compensation, and using models' body images for AI learning.

However, it's necessary to distinguish here. The reports included unsigned contract proposals. The fact that clauses were presented does not mean contracts were finalized with those terms or that images were actually used.

For example, Diesel's casting materials included requests for consent to alter facial features and body proportions or swap faces among participating models. The parent company, OTB, explained that it was for specific campaigns and not representative of standard contracts, stating they are not systematically altering models' appearances with AI.

Calvin Klein's contract proposal confirmed content allowing the use of images, excluding identifiable elements like faces, scars, and tattoos, for AI learning and adjustments without additional compensation. The company stated they are progressing with trials and learning in a rapidly changing field, guided by regulations, industry standards, and respect for models' rights.

L’Oréal's contract included wording about AI use for altering product appearance and color, and retouching. However, the company explained that the standard AI editing tools they use do not affect the model's appearance.

The responses of each company cannot be generalized. Nonetheless, it is clear that new issues are entering contracts that were previously centered on "where and until when photos will be published," now questioning "to what extent can a person's image be altered, learned from, and reused."

The allure of reducing production costs by 90%—whose income is being reduced?

The reason companies have high expectations for AI is clear: it significantly reduces the time and cost required for shooting.

In e-commerce, where new products are continuously launched, there is a burden of coordinating schedules for models, photographers, hair and makeup artists, stylists, and studios to create images each time. If AI can streamline part of the process, the speed of preparing product images can be increased.

According to Reuters, Scayle Studios, operated by About You under Zalando, claims it can reduce production costs by over 90% using AI models for image production. A spokesperson stated that they have reached 130 brands as clients and generated 190,000 outfits in the past month alone.

This "over 90%" is the reduction rate claimed by the service side, not a figure confirmed to have the same effect in all shootings. Nevertheless, it is a sufficiently large number to influence corporate decision-making.

The same article reports that Zalando plans to close two shooting studios in Berlin in 2025, and About You has also closed a studio. Zalando has reduced over 340 staff, with some reassigned to different roles. However, it cannot be concluded that AI was the sole cause of all these staff reductions.

When shooting decreases, it's not only the people appearing on screen who are affected. Those who prepare costumes, create lighting, and finish skin and hair also experience changes in their work.

What appears as "production cost reduction" in company books translates to income reduction for multiple job types on the ground. Considering AI productivity requires looking at both sides.

The backlash on social media is not just because the images are "unnatural"

Consumers are not unconditionally welcoming the changes brought by AI.

One example is the reaction to AI-generated images used by Gucci for a fashion show announcement in February 2026. The reactions introduced here are not directly from the October Reuters article but are related AI advertisement social media posts introduced by news outlets.

In a post on X covered by Fast Company, questions were raised about whether brands using AI-generated ads can be considered luxury and why companies with financial resources don't conduct actual shoots.

The criticism centers not only on the quality of the images.

Luxury brands have conveyed value through craftsmanship, dedication to materials, creativity, and time-consuming production. If the impression of labor-saving is strong in their advertising, consumers may feel a disconnect with the narrative supporting the product's price. This reaction can be seen as questioning the consistency between the brand's explanation and production methods.

On the other hand, BBC's report also introduced positive social media reactions, stating that the images successfully expressed the glamorous Milanese style typical of Gucci.

Therefore, summarizing as "everyone on social media is opposed" is not accurate. These are only some of the posts that could be confirmed and do not represent a survey of the overall consumer opinion.

In an interview with CNN, Gucci's designer, Demna, also expressed a stance of viewing AI as a tool for visualizing ideas. What creators see as a new means of expression may be perceived by the audience as a devaluation of value. This discrepancy fuels the debate.

Does indicating "AI use" solve the problem?

Indicating that something is AI-generated helps viewers understand the origin of the image. However, it doesn't solve everything.

The indication "Created with AI" mainly answers questions about the production method.

Did the person it was based on consent? Was compensation paid? Has it been repurposed for uses different from the initially approved advertisement? These rights issues remain separate from the indication.

Furthermore, what is required changes depending on the purpose of the advertisement.

In image ads conveying a brand's worldview, unrealistic expressions can be appealing. On the other hand, for product images used in purchase decisions, accurately conveying the shape and texture of clothing is crucial.

Creating fantastical and beautiful advertisements and providing reliable information about the actual clothes cannot be measured by the same evaluation criteria.

"Does that clothing really look like that?"—a concern even in Japanese online shopping

This issue is not a distant industry matter for Japanese readers either.

When buying clothes online, wearing images serve as a substitute for trying them on. They help interpret things not discernible from size charts, such as the way fabric falls, sleeve puffiness, and the shape when sitting.

If AI-generated images of clothing create lengths or drapes different from the actual item, what would consumers be basing their purchase decisions on?

This does not mean that errors in product images of specific companies were confirmed in this report. It is a point that Japanese businesses should consider when using AI images for sales.

It's not enough for the generated person to be attractive. There needs to be a process to ensure the color, shape, and texture of the product itself are maintained.

For example, methods such as providing actual front and back photos, close-ups of the fabric, and actual measurements alongside AI-generated images for presentation can be considered. If actual wearing photos are available, the model's height and size worn can also be useful for judgment.

Even if image production costs are reduced, if disappointment arises from differences with the delivered product, it could lead to returns and customer attrition. The success of efficiency improvements does not conclude at the production site alone.

New York's approach of "not processing consent collectively"

An example of a system concerning digital duplicates of models is New York State's Fashion Workers Act.

According to the state's Department of Labor, the main obligations, excluding registration-related ones, have been applied since June 19, 2025. Creating or using digital duplicates of models requires clear written consent, specifying the scope of use, purpose, compensation amount, and duration.

Consent obtained by model agencies must be handled separately from affiliation or agency contracts. The department's FAQ also explains that if a new campaign is not covered by previous consent, fresh approval is required.

This is a system in New York State and does not directly apply to Japanese cases. However, it can serve as a reference for considering contracts.

Separating consent for shooting, photo publication, AI learning, digital double creation, and future advertisement reuse helps reduce misunderstandings between personal understanding and company operations.

In Japanese production sites, at least specifying "what is being created," "where and until when it will be used," "how additional use compensation will be handled," and "who will approve it" will likely be a practical starting point.

Will there be a share for humans in a future where digital doubles work?

Digital doubles have the potential to enable work without the person having to travel or be tied up in shooting. If the person can make decisions each time it's used and is appropriately compensated, it could become a new source of income.

However, if broad use is allowed from the initial shoot and thereafter the person has no involvement or additional compensation, the same technology could work towards taking away future jobs.

Whether it becomes one or the other is influenced not only by the performance of image generation but also by the content of contracts, opportunities for negotiation, and how the profits from efficiency improvements are shared.

Brands have another choice: conveying the process of creation involving humans as a value. Who wore the clothes, who took the photos, and what ingenuity went into creating that one image. As generated images increase, there will likely be people who find meaning in such backgrounds.

AI will enable fashion images to be created faster and in greater quantities than ever before. At that time, the question is not only whether the person in the image is real.

Do the people who made that image have the right to choose and receive compensation? And are the people buying products based on that image receiving reliable information?

The AI transformation of fashion has already entered a stage where these two questions are simultaneously being asked.

Sources and References

1. iTnews/Reuters: Published on October 27, 2026. Interviews with model agencies, shooting contract proposals, company responses, and Scayle Studios' claims of production cost reduction.
   https://www.itnews.com.au/news/fashions-ai-era-face-swaps-and-digital-clones-629941iTnews

2. Fast Company: Reactions on X to Gucci's AI advertisement. Published on February 24, 2026. Introduces posts questioning the relationship between luxury brand value and AI advertising. The content of the posts is summarized in Japanese in the main text.
   https://www.fastcompany.com/91498073/gucci-just-proved-why-luxury-brands-shouldnt-use-ai-generated-ads-imagesFast Company

3. AOL version of BBC article: Criticism and positive reactions to Gucci's AI advertisement. Published on February 25, 2026. Referenced to confirm that social media reactions are not uniform.
   https://www.aol.com/news/gucci-criticised-ai-slop-images-105545561.htmlAOL

4. KESQ version of CNN article: Explanation by Demna on AI usage. Published on February 27, 2026. References Demna's statement viewing AI as a tool for visualizing ideas.
   https://kesq.com/entertainment/cnn-style/2026/02/27/gucci-asks-can-ai-ever-be-luxury/KESQ

5. New York State Department of Labor: Overview and obligations of the Fashion Workers Act. Official information on the effective date, written consent for digital duplicates, and clarity on scope, purpose, compensation, and duration.
   https://dol.ny.gov/fashion
   https://dol.ny.gov/node/58876Department of Labor

6. New York State Department of Labor: FAQ on the Fashion Workers Act. Explains that new campaigns not covered by previous consent require fresh approval.
   https://dol.ny.gov/node/58831