The Authenticity and Purchase Risks of AI Models 'Trying On' Products
-
Petite faces, flawless skin, perfect body proportions—models complement their outfits, making the clothing appear more fitted and trendy... This is a product image found on an e-commerce platform. Upon closer inspection, however, the models 'trying on' the clothes don’t seem to be real people. After messaging the store, the seller readily admitted that the promotional images were generated by AI, 'to better showcase the details of the clothing.'
A recent investigation by Legal Daily revealed that many merchants on multiple e-commerce platforms have adopted AI models to display clothing products. Moreover, some stores on these platforms have begun offering AI model image creation services, such as 'AI-generated model images' and 'replacing real models with virtual ones.'
Are the product effects presented in AI model 'try-on' images reliable, and will consumers accept them? Can AI model 'try-on' images be freely created and used, and what legal risks might they entail? With these questions in mind, the reporter conducted an investigation.
Noticeable Differences in Details May Constitute False Advertising
In AI model 'try-on' promotional images, the clothes fit the models perfectly, with almost no wrinkles. But is the actual product really like that?
Recently, the reporter searched for keywords like 'stylish sets' on major e-commerce platforms and found many products with clearly AI-generated model 'try-on' images. These AI models typically feature exaggerated curves, slender limbs, and extremely refined, cartoon-like facial features, often set against blurred, solid-colored backgrounds.
For example, a store with 6,000 followers on one platform had a set ranked second in its popularity list, with over 450 sales. The product image featured a blonde, blue-eyed cartoon-like model, with both the outfit and the model appearing softened, resembling a comic illustration at first glance.
After purchasing the product, the reporter found significant discrepancies between the received item and the promotional image: the color was off, there were unexpected wrinkles, and the neckline height didn’t match. Even more frustrating, the sleeves in the model image were lantern sleeves ending just above the elbows, while the actual sleeves extended past the elbows and couldn’t be tightened.
The reporter’s experience was not an isolated case. In another online store, a garment advertised with AI model images and selling hundreds of units had reviews like: 'Probably because it’s an AI image, it looks fitted in the picture, but it’s way too loose in reality,' 'The dress is full of wrinkles,' and 'The color of the clothes doesn’t match the picture.'
On social media, under the hashtag 'AI models,' many netizens explicitly stated they wouldn’t buy products advertised with AI models, arguing that such displays increase the risk of online shopping disappointments. Some commented: 'If they don’t even use real models, who would buy it?' 'AI models don’t match human body proportions—how can they be a reference?' 'After AI generates the model images, even the physical product feels virtual. How can authenticity be guaranteed?' Others were more direct: 'I won’t buy any product with AI-generated images.'
Sun Ying, a professor at China University of Political Science and Law and Deputy Secretary-General of the China Consumer Protection Law Research Association, noted that AI applications in e-commerce sales are best suited for real models or hosts using AI to automatically switch between different outfits for display. 'Currently, AI-generated model images still face technical limitations, resulting in discrepancies in details that make it hard for consumers to judge the actual appearance of the clothing.'
AI-generated model images, which cannot fully and accurately represent the appearance, style, color, and patterns of products, inevitably infringe on consumers' rights to information and choice. They violate the obligation of businesses to provide comprehensive, truthful, and accurate product or service information to consumers, and may even constitute false or misleading advertising," said Sun Ying.
Low Production Costs and High Efficiency, but Technical Standards Need Improvement
Why do some businesses prefer using AI-generated model images for product promotion? Industry insiders reveal that this is primarily to reduce photography costs. Reports indicate that after replacing human models with AI models, some businesses saved approximately 40,000 yuan in monthly costs, including expenses for models, photographers, venues, and equipment.
"Many small and medium-sized businesses cannot afford professional models or high-quality photography. AI models help them achieve high-quality product displays at a lower cost. Additionally, for specific shoots requiring foreign models, AI models offer a more convenient and faster solution compared to finding a suitable model in reality," the insider added.
With the growing demand, more software offering AIGC (AI Generated Content) capabilities has emerged. These tools generate AI model images primarily through text descriptions, text combined with images, or image-to-image conversion, enabling one-click generation of AI model "try-on" images.
Investigations reveal that as businesses increasingly seek AI-generated model images, services specializing in creating such images have become a key product for some vendors. On e-commerce platforms, searches for "AI model generation" or "AI clothing model services" yield numerous related products.
In the product descriptions, services include converting mannequin images into realistic models, generating model try-on images directly from clothing photos, and customizing model appearances. Vendors typically use photo-to-image generation, advising, "Provide clothing photos, preferably on mannequins or real models for better results," or "Best materials: model images (mannequin/real model), multiple angles and poses."
Price-wise, inquiries with multiple vendors show that generating 3 to 10 images costs between 100 and 300 yuan. E-commerce operators can also purchase software and tutorials to generate model images independently using AI tools.
One vendor with over 4,000 monthly sales offered a trial service for less than 1 yuan. Following the vendor's instructions, the reporter provided clothing images and keywords like "curly-haired girl," "cute," and "slim," receiving four generated images within minutes.
The models in these images matched the descriptions and appeared highly realistic, almost indistinguishable from real people. However, the clothing details—such as neckline design, sleeve length, hemming, presence of belts, and accessory placement—differed significantly from the original images, aside from color.
"You can refine the keywords for better results," the vendor suggested in response to the reporter's concerns, adding that providing images of real or mannequin models would yield more satisfactory outcomes.
In the view of Zheng Zhifeng, associate professor at the Civil and Commercial Law School of Southwest University of Political Science and Law and deputy director of the Cyberspace Governance Research Institute, the use of AI technology to generate model images reflects specific application scenarios of AI, which helps reduce sales costs. At the same time, generative AI technology can efficiently produce various model images in bulk, many of which are highly creative and aesthetically pleasing. However, the generated content is unlikely to perfectly match real models, and its authenticity and accuracy require further algorithmic advancements.
Merchants Must Fulfill Labeling Obligations and Beware of Portrait Infringement Risks
To regulate the application of AI-generated content, the Cyberspace Administration of China (CAC) and two other departments reviewed and passed the Regulations on the Management of Deep Synthesis in Internet Information Services, which took effect on January 10 this year. The regulations explicitly state that deep synthesis service providers and technical supporters must strengthen technical management, regularly review, evaluate, and verify the mechanisms of generative algorithms. If they provide models or tools capable of generating or editing biometric information such as faces or voices, they must conduct safety assessments either independently or through professional agencies in accordance with the law.
The regulations also specify that deep synthesis service providers offering services like face generation, face replacement, face manipulation, or posture manipulation—which may lead to public confusion or misidentification—must prominently label the generated or edited content to inform the public of its AI-generated nature.
The Interim Measures for the Management of Generative Artificial Intelligence Services, reviewed and passed by the CAC and six other departments, will take effect on August 15 this year. It reiterates that generative AI service providers must label images, videos, and other generated content in accordance with the Regulations on the Management of Deep Synthesis in Internet Information Services.
"In the e-commerce sector, particularly in product and service displays, the use of AI is an inevitable trend," said Liu Jinrui, a researcher at the China Law Society's Institute of Rule of Law. He noted that merchants using AI for product displays must comply with the aforementioned regulations, clearly informing consumers that the model images are AI-generated to fulfill labeling obligations. Additionally, they must adhere to special regulatory requirements under laws such as the Advertising Law, E-Commerce Law, and Consumer Rights Protection Law.
However, an investigation revealed that neither merchants using obvious AI-generated cartoon-style model images nor those advertising "highly realistic AI models" on social media platforms labeled their product pages to indicate the use of AI-generated images.
"Platforms bear supervisory responsibilities. If they also provide technical services, they must comply with laws like the Interim Measures for the Management of Generative Artificial Intelligence Services and take effective measures to oversee AI-based promotional activities to maintain platform order," Zheng Zhifeng said.
Notably, since some AI databases contain vast amounts of facial data, AI-generated model images may closely resemble real human faces. This has raised concerns among netizens about potential copyright infringement.
Wang Yegang, associate professor and doctoral supervisor at the Central University of Finance and Economics Law School, stated, "If AI-generated images are highly similar to someone's portrait to the extent that the person can be identified, such images should be considered as that person's portrait. Merchants must obtain consent from the portrait rights holder before creating or using such images; otherwise, they may infringe on the individual's portrait rights."
Zheng Zhifeng emphasized that merchants must avoid using AI technology for illegal purposes, such as intentionally inputting prohibited keywords to generate inappropriate content or content that infringes on others' portrait rights, reputation rights, or personal information rights. Even for legitimate uses, merchants must exercise due diligence to ensure compliance and report any suspicious content to service providers promptly.