Last week we published three early findings. Today the full report is out: nine findings from 331 fashion professionals on how the industry actually produces, publishes, discloses and governs AI-generated imagery, two months into the era of mandatory transparency. The complete report, with every chart, question wording and cut, is free to download. This post is the open summary.
Methodology, first
Fieldwork ran September 19 to 25, 2026. 331 professionals completed the survey: brand and marketing leads (28%), creatives and photographers (26%), founders and executives (18%), e-commerce managers (12%) and compliance and legal roles (8%), from solo practitioners to organisations above 1,000 people, across 29 countries. Topline figures carry a margin of error of ±5 points at 95% confidence (±6 on the AI-user base). Respondents skew AI-active, reached through the ecosystem of an AI imagery platform; we treat that as the report's frame rather than its footnote. These are the numbers from the population already working with AI content, where the rules bite first, which makes the gaps below floors, not ceilings. No finding in this report cuts any segment below 50 respondents; full question wording and methodology ship inside the PDF.
The nine findings
1. Scale: AI imagery is production infrastructure, not experiment
67% of organisations use AI-generated or AI-edited imagery in commercial content. Among users, two-thirds say more than 10% of their published imagery is already fully AI-generated, and one in four say more than a quarter. 81% expect the share to grow in the next 12 months, 42% significantly. Whatever the industry debate sounds like, the production floor has decided.
2. The disclosure gap (from the early findings)
67% use AI imagery; 16% always label it. The full report adds the channel view: disclosure practice varies by where the imagery is published, which suggests labelling is treated as a platform-policy question rather than the legal duty it now is.
3. Readiness: the letter most companies cannot answer
Asked whether they could answer a regulator or business partner today, documented and ready, 15% said yes. 32% could answer partially, needing time to assemble it, and 32% could not answer today. A further 35% do not know whether the rules apply to them at all.
4. The fines nobody has heard of (from the early findings)
71% of respondents did not know that breaches can carry fines of up to 15 million euros or 3 percent of worldwide turnover. The budget finding below is the natural consequence.
5. The awareness gap: C2PA
Before taking the survey, 50% of respondents had never heard of C2PA or Content Credentials, and only 20% both knew the name and understood what it does. The standard the industry's infrastructure is converging on is, for most of the people publishing the images, still invisible. If you are in that majority, the ten-minute version is on our blog here.
6. Money: the budget void and the cost dividend
34% of organisations allocate no budget to content authenticity or provenance tooling, and only 22% have a dedicated budget today. Meanwhile, among AI adopters, 66% report per-image production costs reduced, 28% by more than half. The report's uncomfortable juxtaposition: the technology is delivering measurable savings, and almost none of the dividend is being reinvested in the compliance the same technology now requires.
7. The procurement shift
63% of respondents already require or intend to require content provenance from their creative vendors and image suppliers. Enforcement authorities move slowly; procurement moves quarterly. If this number holds, the practical enforcer of provenance standards in fashion will be the RFP, not the regulator. The vendor checklist we published this week is built for exactly that conversation.
8. The 2028 consensus
66% believe disclosing AI-generated imagery will be standard practice in fashion by 2028, whether by law or by market expectation; only 12% believe it will not happen. The industry, in other words, already believes in the destination. The findings above measure how far behind the belief the practice runs.
9. The exclusive: duties versus practice
This is the analysis no other survey can produce, because it requires mapping survey answers onto the legal triage of image operations. We asked what respondents actually do with AI (synthetic models, virtual try-on, background generation, generative retouching) and matched each operation against the marking and labelling duties it triggers.
The result: 95% of AI-using organisations perform at least one operation that legally requires machine-readable marking, and 66% perform operations in the deepfake class that require a visible label. Among that last group, 51% never label. That triangle, what companies do, what the law requires for it, and what companies actually disclose, is the single chart we expect to be cited most, and it is chart one in the PDF.
What the report recommends
The full PDF closes with a short, unglamorous programme: name an owner for AI-content compliance; classify your imagery against the operations triage; put visible labels into the publishing workflow for deepfake-class content; ask your vendors the ten questions; and keep export records you can produce. None of it requires budget approval to start, which, given finding six, is the point.
For journalists and researchers
The report is free and ungated for citation purposes on request, the chart pack is available in ready-to-publish formats, and every chart carries its n, field dates and source line. Cite as: Modelia, The State of AI-Generated Content in Fashion 2026, October 2026. For the underlying question wording, cut rules or interviews, write to press@modelia.ai. Download the full report here.
These findings come from The State of AI-Generated Content in Fashion 2026, free, with full methodology and every chart. And anything you read here can be tested: check any image in our free public verifier.
