TikTok is testing a system that searches for AI replicas of creators
TikTok has begun testing an opt-in likeness-detection tool with a limited group of creators in the United States. Participating creators verify their identity through a live selfie scan and an identity-document check performed by verification provider Jumio. TikTok then scans for AI-generated content that may use the creator’s likeness, allowing the creator to review suspected matches and report unauthorised posts or accounts. TikTok says it does not retain identity documents and uses facial information for likeness matching and identifying potential misuse.
Why visual creators should care
Most platform remedies begin only after a creator happens to discover a deepfake. TikTok’s pilot moves towards continuous, identity-based detection. That could make it easier to locate synthetic endorsements, impersonation accounts and harmful generated images, but it also requires creators to provide additional biometric information to activate the protection.
Practical takeaway
Decide in advance who will manage likeness monitoring for a creator, model or represented performer. Preserve screenshots, URLs, account names, timestamps and the genuine source photographs connected to each suspected match. Before completing identity verification, review which company processes the selfie and identity document, what data is retained and who can access the results.
What to watch next
Watch whether TikTok expands the pilot beyond selected US creators, how accurately it distinguishes legitimate fan work and authorised campaigns from misuse, whether reporting leads to rapid removal and whether agencies can manage detection on behalf of represented talent.
Sources
Synthetic product demonstrations are spreading through TikTok Shop
AI-generated shopping videos featuring virtual presenters and synthetic product demonstrations are becoming increasingly common on TikTok Shop. Some affiliate creators use avatars to produce promotional content at scale, while brands have limited control over who markets their products through TikTok’s open affiliate system. Some companies, including SharkNinja, have reportedly prohibited AI-generated affiliate content.
Why visual creators should care
A synthetic presenter can appear to hold, wear or test a product that was never physically used. That creates risks for commercial photographers, filmmakers and brands because audiences may interpret a generated demonstration as real product experience, while authorised campaign imagery competes with large volumes of cheaper synthetic promotion.
Practical takeaway
Brands should state whether affiliates may use AI-generated people, product demonstrations or altered pack shots. Agreements should require disclosure, accurate product representation and removal rights when synthetic content misstates size, colour, performance or use.
What to watch next
Watch whether TikTok strengthens enforcement of AI labels in shopping content, gives brands greater control over synthetic affiliates or distinguishes between AI-assisted editing and entirely generated product demonstrations.
Sources
Clearview AI’s facial-image settlement has been overturned on appeal
The Seventh US Circuit Court of Appeals has overturned a proposed class-action settlement concerning Clearview AI’s collection of facial images from the internet. Clearview built its facial-recognition database by scraping photographs from websites and social platforms. The rejected settlement would have given class members a financial interest connected to the company rather than a conventional cash payment.
Why visual creators should care
The underlying dispute concerns the transformation of publicly accessible photographs into biometric identifiers. A photograph posted for editorial, portfolio or social use can acquire an entirely different value and risk profile when it becomes searchable facial-recognition data.
Practical takeaway
Photographers working with identifiable people should avoid promising that publication controls every later technical use of an image. Model releases, privacy notices and client guidance should distinguish ordinary publication from facial recognition, biometric analysis and AI training.
What to watch next
Watch whether the parties negotiate a replacement settlement, whether litigation resumes and whether the final remedy addresses deletion or restriction of facial data rather than only financial compensation.
Sources
The authenticity problem may begin inside the camera
Researchers examining generative processing inside camera image pipelines warn that functions such as AI-powered digital zoom and low-light enhancement can add details that were not recorded directly by the sensor. They propose storing enough information in ordinary JPEG or HEIC metadata to recover a version of the image from before the hallucinated detail was introduced.
Why visual creators should care
Professional authenticity rules often treat a camera-original file as the strongest available evidence. That assumption becomes less reliable when generative reconstruction occurs during capture and is embedded in the first file the photographer receives.
Practical takeaway
Photographers working in documentary, evidentiary or competition settings should identify which in-camera modes reconstruct detail. Where authenticity matters, preserve RAW sensor data when available and record whether digital zoom, computational low-light processing or generative enhancement was enabled.
What to watch next
Watch whether camera manufacturers expose clearer capture-time AI disclosures, whether provenance standards record generative camera processing and whether competitions revise their definition of a camera original.
Sources
A detector’s explanation may matter as much as its final score
Researchers developed and tested AI-image detectors alongside 16 explainability methods intended to show why a system classified an image as synthetic. The work used a large photorealistic image dataset and responses from 100 participants to examine whether machine-generated explanations aligned with visual cues people could understand.
Why visual creators should care
A publication, competition or agency cannot responsibly accuse a creator of submitting synthetic work merely because a system produces a high probability score. Reviewers need to know which parts of the image influenced the result and whether those features could also occur through compression, retouching or ordinary camera processing.
Practical takeaway
When evaluating detector software, ask whether it identifies the visual evidence behind its result, whether the explanation survives resizing and compression and whether a qualified reviewer can challenge the conclusion.
What to watch next
Watch whether commercial detection systems begin providing interpretable evidence, uncertainty ranges and transformation-specific results instead of a single synthetic-or-real classification.
Sources
People judge synthetic portraits differently depending on age and viewing device
A study involving 1,664 participants tested whether people could distinguish real portrait photographs from images generated using ChatGPT-4o and Imagen 3. Average accuracy was 85.2 percent, but performance varied by age, device and confidence. Participants using computers performed better than mobile users, and accuracy declined with age, particularly on mobile devices.
Why visual creators should care
Authenticity judgments are often made quickly on phones, social platforms and compressed previews. The study suggests that an image’s apparent credibility is partly determined by who is looking, on which device and under what conditions—not only by the quality of the generated portrait.
Practical takeaway
Do not ask editors, clients or competition judges to authenticate portraits from small mobile previews. Review the highest-quality available file, inspect source evidence and avoid treating personal confidence as proof of accuracy.
What to watch next
Watch whether later studies reproduce the results using newer image generators and whether platforms account for mobile viewing conditions when designing labels and authenticity interfaces.
Sources
