Likeness rights, model releases, and the legal framework around AI-generated imagery are the most under-discussed operational risk in fashion ecommerce right now. Most brands adopting AI catalog production have not updated their model release templates, their content licensing terms, or their internal QC checks for AI-generated content. The case law is still developing in most jurisdictions, but the prudent practices are already clear. This is the practical 2026 guide to likeness, releases, and AI imagery for fashion brands. This article is general information, not legal advice for your specific situation.
What “likeness rights” actually protect
Likeness rights protect a person's right to control the commercial use of their identifiable image, including face, body, voice, and other distinctive features. The legal frameworks vary by jurisdiction. In the United States, publicity and privacy rights differ by state. In the European Union, GDPR, national personality rights, advertising rules, and contract law can all be relevant. In the United Kingdom, there is no single standalone image right; data protection, passing off, misuse of private information, contract, and intellectual-property rules may apply depending on the facts. The practical rule for commercial catalog production is to obtain clear permission rather than assume that one doctrine works everywhere.
For AI-generated imagery, the question is whether the generated image “contains” an identifiable person's likeness. If the AI was trained on or prompted with a specific real person's image, and the output is recognisably that person, the likeness question applies. If the AI generated a synthetic person not corresponding to any real individual, the likeness question does not apply in the same way — though other legal concerns can still apply (consumer protection, deceptive advertising, etc.).
The three-layer framework for AI fashion imagery
Practically, fashion brands are working with three different layers of AI-generated person imagery, each with different consent requirements. Layer one: fully synthetic platform models — the AI generates a person who does not correspond to any real individual. Layer two: AI-generated images that started from a real reference photograph (the brand's own model, a real creator, a stock-licensed person). Layer three: AI-generated likeness of a public figure or another real person not contracted by the brand.
The risk gradient is clear. Layer one carries the lowest likeness risk. Layer two requires the consent chain back to the original real person to be intact and commercial-use-permissive. Layer three is the highest risk and should generally be avoided entirely.
How the Apiway creator marketplace handles the consent chain
Apiway's creator marketplace operates explicitly inside layer two. Creators upload photo sets of themselves with terms permitting commercial use of derivative AI-generated imagery by purchasing brands. The consent chain is documented at the source. Brands using the marketplace have a clean licensing story that traces back to the creator's explicit consent. This is one of the meaningful legal advantages of the marketplace approach over scraping web imagery or using AI tools whose training data provenance is opaque.
For brands using their own real model imagery as input to AI tools (uploading the brand's own model photograph as a reference for the AI to render the catalog), the model release contract has to explicitly authorise AI-generated derivative use. Most pre-2024 model releases did not contemplate this and should be updated. Brands shipping AI catalog content from older model agreements should review with counsel.
Updating the model release template for AI use
A modern fashion model release for the AI era explicitly addresses several points that older templates did not. First, authorisation for the model's likeness to be used as input to AI tools and for the AI tools to produce derivative imagery. Second, the scope of the derivative imagery the brand may produce — same garment, different garment, same model identity, modified model identity. Third, the commercial channels and territories the AI-derived imagery can be used in. Fourth, duration and termination terms, including whether AI-derived imagery can continue to be used after the underlying contract ends.
Brands operating across jurisdictions need to verify the release is enforceable in each jurisdiction the model resides in or where the imagery will be published. GDPR or UK GDPR considerations may apply when identifiable people's personal data is processed in the EU or UK. California's right-of-publicity statute separately addresses knowing commercial use of another person's name, photograph, or likeness without prior consent. Which privacy statutes apply depends on the business, processing activity, thresholds, and people involved — not merely on publishing an image to a large audience. The release template is the operational anchor for all of this; investing in a properly updated template pays back across every shoot.
AI imagery disclosure and consumer protection
Disclosure duties depend on the content, the system, and the jurisdiction. Article 50 of the EU AI Act applies from 2 August 2026. Among other duties, providers of systems that generate synthetic content must support machine-readable marking, and deployers must clearly disclose AI-generated or manipulated image, audio, or video that qualifies as a deepfake. Evidently artistic or fictional works receive a narrower disclosure rule; ordinary editing is not automatically a deepfake. California also requires certain large generative-AI providers to support manifest and latent disclosures for qualifying content. These rules do not make every AI-assisted fashion image subject to the same label, so brands should classify each workflow with counsel and build flexible disclosure and provenance fields into the catalog system.
Training data and model provenance questions
A separate legal layer concerns the AI tool's own training data. If the tool was trained on imagery without explicit licensing for that training, the downstream brand may face provenance, contract, or reputation risk. Vendor practices differ, and broad claims about licensed or synthetic training data should not be accepted without evidence. Brands should verify the tool's training practices, contractual protections, retention terms, and treatment of customer uploads before adoption. Apiway's position is that we never train on user uploads; the training data is curated separately and the brand's own catalog inputs stay private.
Practical checklist for fashion brands
First, audit the existing model release template and update for AI-derivative use with counsel. Second, document the consent chain for every AI imagery source used in production catalogs. Third, build a disclosure capability into the catalog system for AI-generated imagery. Fourth, verify the AI tool vendor's training-data position before adoption. Fifth, track emerging AI-content disclosure regulation per jurisdiction the brand publishes in.
None of this is paralysing operational risk; it is normal commercial discipline applied to a new technology. Brands that build the discipline early are insulated from the inevitable regulatory and case- law evolution; brands that defer the discipline accumulate technical debt that is expensive to unwind.
Related reading
See our deep-dive on photo set licensing for AI commercial use, our essay on training-data privacy, and the full Apiway blog for more legal and operational guides.
Primary references: the official EU AI Act, Article 50, the European Commission's Article 50 transparency guidance, California Business and Professions Code chapter 25, and the UK ICO's AI and data-protection guidance.