Fashion After the Algorithm

In July 2025, H&M released its first campaign images featuring digital twins of real models. The women shown in the images had not posed for those particular scenes in a conventional studio. Their professional appearances had been translated into digital counterparts that could be placed against the backdrops of several fashion capitals. The company presented the project as a transparent collaboration among models, photographers, a creative team and generative technology.

The images mattered less for their visible aesthetic than for the economic order forming behind them. A model is ordinarily hired for a defined date, campaign, territory, period of use and set of media. A digital twin, by contrast, can be reused after the original commission, altered, redressed, aged, rejuvenated or placed in an environment that never existed. The central question of AI fashion is therefore not whether an image is real or artificial. It is who retains the authority to decide how bodies, style, labor and reuse are governed inside a synthetic image economy.

Artificial intelligence in fashion has already moved far beyond faster sketching. It searches archives, generates design variants, visualizes textiles, models fit, forecasts demand, personalizes product recommendations and produces campaign material. Considerable economic pressure sits behind this development. For The State of Fashion 2026, McKinsey and The Business of Fashion surveyed executives and consumers across several major markets. Ninety-two percent of the companies surveyed intended to increase their investment in artificial intelligence. Only one percent described their use of it as mature. At the same time, 46 percent of executives expected economic conditions to deteriorate, while only 25 percent anticipated improvement.

These figures describe more than technological ambition. They reveal an industry investing faster than it is developing shared rules for quality, liability, authorship and labor. Within that gap, the future character of AI is being decided. It may become a sophisticated instrument of human creativity, or an infrastructure for the quiet erosion of creative value.

From Tool to Cultural Decision

Fashion is not simply the production of wearable objects. It is a system of signs. Clothing can express belonging, distance, religion, class, gender, profession, resistance or memory. A design acquires meaning not through silhouette alone, but through the cultural decisions embedded in material, proportion, reference, manufacture and presentation.

As long as artificial intelligence remains a controlled tool, human responsibility is comparatively clear. A designer may use a system to visualize alternatives, test a textile combination or execute a local adjustment more quickly. Current research into human-led generative fashion processes describes precisely this form of collaboration: the human defines pose, structure, modification and design conditions, while the system develops variants consistently. The technology enlarges the field of possible designs, but it does not replace the judgment required to determine which form can be defended culturally, materially and aesthetically.

The situation becomes more difficult when a system does not merely execute, but selects. Trend forecasting and recommendation models are built on historical data. They register what has been purchased, clicked, saved and shared, then calculate which color, length or silhouette is most likely to succeed.

Probability, however, is not the same as relevance. A system trained on existing popularity can become highly efficient at offering what has already been recognized. It may treat deviation as risk, even though the history of fashion often begins where a person refuses the expected solution. A design for which no one has yet searched is harder to predict than another variation within a familiar micro-aesthetic.

This creates a fundamental conflict. Fashion businesses want AI to reduce uncertainty, while creative renewal often begins with a decision that cannot be inferred from the past.

Personalization Is Not Automatically Individuality

According to McKinsey, the number of shopping-related queries on generative AI platforms rose by 4,700 percent between 2024 and 2025. Consumers increasingly discover fashion not only through search engines, shops or social networks. They formulate complex requests: a dress for a particular occasion, suited to body, budget, weather, cultural setting and personal aesthetic. A system can turn those conditions into a selection or a bespoke visual proposal.

This promises a new degree of personal precision. Personalization can help address sizes, body forms, disabilities, religious requirements and gender identities that have long been underrepresented in standardized ranges. It can make overwhelming inventories easier to navigate and help smaller labels become discoverable.

It can also narrow a person's taste into an increasingly restrictive profile. Someone who repeatedly selects black, clean lines and a limited group of brands is shown more of the same. What appears to be personal fit can become an elegant mechanism of repetition. The system does not merely confirm existing taste. It begins to shape taste by making alternatives less visible.

Individuality means more than a machine knowing what someone is likely to prefer. It includes the possibility of surprise, change and a decision that does not emerge from one's own data history. A culturally intelligent recommendation system should therefore do more than optimize similarity. It should permit a controlled encounter with difference.

The Digital Body as an Economic Asset

New York's Fashion Workers Act has been in force since 19 June 2025 and explicitly treats digital replicas as a labor issue. Model management companies and clients must obtain separate, clear written consent before creating or using a digital replica. The consent must state purpose, scope, compensation and duration of use. A general power of attorney cannot simply absorb authority over a digital replica.

The provision is significant because it does not treat a digital likeness as ordinary retouching. Face, body and voice are recognized as professionally exploitable assets. The law does not resolve every international question, but it establishes a sensible minimum principle: consent to a photo shoot is not automatically consent to a potentially unlimited synthetic future.

A serious agreement must clarify at least:

• which elements of the person will be digitally captured,
• the campaigns, media and territories in which the replica may be used,
• whether face, body, skin, age, voice or movement may be altered,
• whether each new use will be separately compensated,
• whether the replica may be transferred to third parties,
• whether it may be used as training material,
• how long the files will be retained,
• how the person concerned can document and stop unauthorized use.

These questions do not concern models alone. Photographers, stylists, make-up artists, hair artists, set designers and retouchers all contribute to an image language that can later be reconstructed synthetically. A campaign that appears to have been created without a traditional shoot may still depend on a substantial body of earlier human labor. Efficiency must not render that history invisible.

Synthetic Campaigns and the New Duty of Honesty

In July 2024, Mango released a campaign for its Teen line created entirely with generative AI. According to the company, photographs of real garments were used, a generative model was trained, outputs were selected by internal design, art and photography teams, and the chosen images were then refined. The case demonstrates that the distinction is not simply between human work and machine work. Even a synthetically produced campaign may rest on a complex human production chain.

For that reason, a general label such as "AI assisted" is insufficient. It does not reveal whether only a background was changed, a model was fully generated, or a real garment was synthetically placed on a digital body. Good disclosure does not need to explain every technical detail, but it should answer four questions in language a viewer can understand:

• Is the person shown real, digital or hybrid?
• Does the product exist in the form in which it is depicted?
• Which substantial elements of the image were generated or altered?
• Was the result reviewed by an accountable human team?

The transparency obligations in Article 50 of the European Union Artificial Intelligence Act have applied since 2 August 2026. Providers of certain generative systems must support technical measures for machine-readable marking. Those who publish deepfakes or particular forms of AI-generated content may be required to provide visible disclosure. Differentiated rules apply to artistic and editorially controlled material, but the direction is clear: the provenance of synthetic content is becoming part of professional communication.

Fashion houses should not treat this development as a burdensome labeling requirement. In an industry that establishes value through material, origin and craftsmanship, image provenance is another form of the same credibility. A luxury object loses persuasive force when its material precision is promoted through a campaign that leaves viewers unable to determine whether the body, place or even the product existed.

The Sustainability Paradox

Artificial intelligence and three-dimensional design processes are frequently associated with sustainability. Digital samples can reduce the need for physical prototypes. Virtual fit testing can save material, transport and development time. A 2026 study of 3D CAD adoption among small fashion businesses, however, shows that these benefits do not arise automatically. Based on interviews with small design enterprises, the study identifies cost, hardware, training and workflow disruption as substantial barriers. Environmental benefit appears only when digital processes genuinely replace physical samples, unnecessary production or transport.

Faster development can also produce the opposite result. If a company creates more collections, more variants and shorter micro-trends within the same period, total volume rises. A single process becomes more efficient while the system as a whole accelerates.

Sustainability must therefore be measured through absolute results:

• How many physical samples were avoided?
• By how much did actual production volume decline?
• Were returns and write-downs reduced?
• How did energy and computing demand change?
• Was product life extended?

Without such measures, sustainability remains a plausible narrative rather than a demonstrated outcome.

Who Owns Style?

Style is harder to protect legally than a specific work. It develops over years through recurring decisions, cultural influences, material experience and public recognition. That is precisely why it is attractive to generative systems. A model does not need to copy a particular silhouette in order to reproduce the recognizable atmosphere of a designer or house.

The imbalance is particularly acute for independent labels. Large companies can afford legal advice, data management and technical control. A small designer may discover that a distinctive visual language is appearing across mass-generated images without being able to prove which work was used or what financial harm resulted.

WIPO and EUIPO therefore directed their joint 2025 IP Management Clinic specifically towards fashion businesses and small enterprises. Design rights, trade marks, copyright, trade secrets, licensing and digital enforcement are no longer treated as isolated specialist issues, but as parts of a coherent business strategy.

The fashion industry needs traceable training data, licensing models and provenance systems that are not accessible only to major rights holders. Style should not become a freely extractable raw material merely because it does not fit neatly within one legal category.

Human Judgment Becomes More Valuable

The fashion industry will not abandon artificial intelligence. Economic pressure is too strong, the range of applications is too broad, and consumers are already adopting new forms of search and visualization.

The decisive question, however, is not how many processes can be automated. It is which responsibilities must never be automated.

A system can recognize patterns, but it cannot assume responsibility for the cultural meaning of a reference. It can visualize bodies, but it cannot grant informed consent. It can forecast demand, but it cannot decide whether an increasingly narrow prediction of taste is a desirable cultural objective. It can perfect an image, but it cannot guarantee that the product, the person and the labor behind it were treated fairly.

In the age of synthetic imagery, the value of human fashion does not diminish. It becomes more precisely definable. It lies in judgment, material knowledge, consent, responsibility, historical sensitivity and the ability to design something no existing data series anticipated.

The most elegant fashion of the future will not be the fashion that conceals its technology most successfully. It will be the fashion that makes visible where a human being made the decision.

Six Questions Readers May Ask After This Article

1. At what point does AI-assisted fashion design become independent authorship by the system?

Under current legal interpretations, an AI system is not a human author. The decisive issue is which creative choices a person can demonstrate having made. Someone who determines concept, selection, structure, material, modification and final form has a stronger basis for human authorship than someone who enters a broad instruction and adopts the output unchanged. Documentation of the design process therefore becomes increasingly important. Sketches, versions, sources, corrections and selection decisions can show where individual creative work lies. The precise threshold varies by jurisdiction and technical application. Businesses should preserve not only the final image, but the entire chain of decisions.

2. Does hyper-personalized fashion strengthen individual style or narrow it?

Both outcomes are possible. Personalization can account for sizes, body forms, disabilities and cultural needs that standardized ranges have long neglected. It can also create a feedback loop: the system shows more of what has already been clicked or purchased, turning past preferences into an increasingly confined future. Genuine individuality requires not only precision, but surprise. Recommendation models should therefore explain why something is being shown and deliberately offer options beyond an established profile. Otherwise, the result is not personal style, but a high-resolution mirror of previous consumption.

3. Which rights should models retain over their digital twins?

At minimum, control over creation, purpose, duration, territory, alteration, transfer and compensation. Each new use should be clearly connected to the original contract or approved separately. Changes to body, age, skin, voice and sexual presentation are especially sensitive, as are uses in political or reputation-damaging contexts. A model also needs access to evidence showing where and when the replica was used, together with an effective means of stopping unauthorized use. New York's Fashion Workers Act provides an important minimum standard, but it does not replace careful international contracting.

4. Can artificial intelligence genuinely reduce overproduction in fashion?

Only when efficiency results in less physical production. Virtual samples, more accurate sizing models and better demand planning can reduce material use, transport and returns. The benefit disappears if faster development leads to more collections, more variants and shorter trend cycles. Companies must therefore measure absolute outcomes, not simply the efficiency of an individual process. Relevant indicators include the number of samples avoided, actual production volume, returns, write-downs, energy consumption and product life.

5. How can a brand disclose AI use transparently without turning transparency into a marketing performance?

Disclosure should be specific, easy to find and proportionate. It need not explain every technical detail, but it should state whether the model is real or digital, whether the product physically exists, which substantial parts of the image were generated, and whether a human creative team reviewed the result. A brief, standardized provenance notice is more credible than a dramatic campaign about the supposed future of creativity. Transparency is convincing when it is treated with the same normality as material information or an image credit.

6. Which human skills become more valuable as AI expands in fashion?

Above all, skills that are not based on statistical similarity: cultural judgment, material knowledge, craft experience, accountability, empathy, negotiation and historical interpretation. The ability to set limits also becomes more important. A designer who knows when a system is useful and when it dilutes the meaning of a design holds greater creative authority than someone who merely generates the greatest number of variants.

Sources and Further Research

1. McKinsey and The Business of Fashion, The State of Fashion 2026, 18 November 2025.
2. H&M Group, H&M continues its exploration of creativity with AI, 2 July 2025.
3. The Guardian, Fashion models face AI job losses as brands create digital twins, 30 March 2025.
4. Mango Fashion Group, Mango creates the first campaign generated by artificial intelligence for its Teen line, 10 July 2024.
5. New York State Department of Labor, Fashion Workers Act.
6. New York State Department of Labor, Responsibilities of Fashion Management and Clients.
7. European Commission, Transparency obligations under Article 50 of the AI Act.
8. Guopeng Zhang, Controllable Generative Systems for Fashion Prototyping, 2026.
9. Forming digital futures: 3D CAD technology adoption among small fashion designers, 2026.
10. WIPO and EUIPO, IP Management Clinic 2025 - Fashion, 2025.
11. WIPO, The Power of Intellectual Property for Fashion SMEs, 2025.