29.Aug.2026
Fake escort profiles are an old problem. What has changed is how easy it is to invent a face that never existed, or to paste someone else's likeness onto a listing they never approved. Clients can lose deposits; providers can find their photos stolen or morphed.
This guide covers practical checks: Content Credentials (C2PA), Google's "About this image" and SynthID, reverse image search, and visual artifacts — plus the limits of "AI detector" websites.
Scammers use generated or heavily edited images to invent a "provider," impersonate a real one, or dress a borrowed face up as exclusive. A real photo can also be generatively edited or posted out of context. No directory can promise every image online is camera-original, so use layered checks.
For a broader scam checklist, see How to Avoid Getting Scammed by Fake Escorts. For stolen photos that already exist online, start with How to Reverse Image Search Fake Profiles.
The Coalition for Content Provenance and Authenticity (C2PA) publishes an open standard called Content Credentials. C2PA describes them as a kind of "nutrition label" for digital files: signed information about origin, edits, and whether a tool flagged the asset as AI-generated or modified.
C2PA's 2026 deployment guidance says you can check a file at the free public tool verify.contentauthenticity.org. Upload the original exported file. The tool reports whether credentials are present and whether they are valid and/or trusted, plus signer, time, and tool when available.
Important limits, from that same guidance:
Adobe's Inspect help page describes a similar upload-and-inspect workflow and notes that Content Credentials are cryptographically signed and tamper-evident, unlike ordinary editable metadata.
Google Search Help's Learn more about an image explains About this image. When available, it can show:
Google also warns that credit and digital source-type metadata can be modified with generation and editing tools, so metadata alone may not be accurate.
A July 2024 Google blog post adds that About this image can identify a Google DeepMind SynthID watermark if that watermark is present in the pixels, and reminds readers that metadata can be added or removed by whoever posts the image.
DeepMind's SynthID page describes an imperceptible watermark added when supported Google (and some partner) tools generate or alter content. You can ask Gemini, Search or Chrome whether an upload was created or altered by Google AI. In 2025 DeepMind also launched a SynthID Detector portal for watermarks from Google and named partners (it lists OpenAI, NVIDIA, Kakao, and says Apple is coming).
A missed watermark does not prove a photo is human-made. It only means that particular signal was not found — the image may come from a tool that never added SynthID, or the signal may have been degraded.
AI-generated faces that never existed may not match older photos. Stolen or lightly edited photos often will. Run the image through more than one search:
Look for the same person under a different name or city, stock-photo hits, or a date years earlier than the "new" profile claims. Steps: How to Reverse Image Search Fake Profiles.
Image models keep improving, so "tells" that were obvious a year ago may be subtle now. Treat the following as reasons to look closer, not as a verdict. They also appear in real photos (motion blur, unusual jewelry, heavy makeup, aggressive beauty filters).
Hands and anatomy. Extra or merged fingers, odd nails, or a hand that does not attach cleanly to a wrist still appear in some generated shots.
Text, jewelry and accessories. Signs, logos and book spines may dissolve into almost-letters; earrings may not match, or a necklace may melt into skin or change between photos.
Backgrounds, lighting, skin and hair. Watch for broken perspective, repeating patterns, light on a face that does not match the window, uniformly airbrushed pores, or hair that smears into the background.
Over-consistent faces. A set of perfectly lit close-ups with no ordinary phone noise can come from one prompt. A face whose bone structure or eye color shifts across "her" photos is also a warning.
Inconsistency with the rest of the profile. Glamorous hotel suites in every shot, but a story about a quiet local incall; or photos that do not age with the claimed timeline. Cross-check with screening habits described in Escort Screening.
If someone has copied or AI-morphed your images, save the URLs, reverse-search, and report the listing. See also SecretHostess and EuroGirls on copied ads.
Public "is this AI?" score sites are tempting because they give a percentage. They are a weak last step, not a judge.
Researchers commenting to NIST have argued that consistently detecting all AI-generated content is not a solved problem, that detectors are error-prone, and that generators and detectors stay in a catch-up cycle. They also noted that watermarks and metadata can be removed or forged, and that some early consumer detectors were withdrawn because accuracy was too low.
NIST's GenAI evaluations measure how well systems tell fully AI-generated images from human-made ones, and they set aside many real-world cases such as light AI enhancement. Forensic briefings have also flagged poor generalization to new generators and uncertain results after post-processing (the compression and cropping social platforms apply).
Practical rule: a detector "AI" score is not evidence you can take to a bank or a booking decision. A detector "real" score is not a green light. Combine provenance tools, reverse search, profile consistency, and ordinary screening.
For clients:
For providers:
Velvet22 cannot and does not guarantee that every photo on the internet is authentic. On this site, the published rules are:
Those policies reduce some risks. They are not a substitute for the checks above, and they are not a promise that a meeting will be safe.
If Content Credentials are missing, is the photo fake?
No. C2PA's guidance says credentials may never have been added, or may have been stripped by a screenshot or re-export.
Does "About this image" always label AI photos?
No. It shows context when Google has it. SynthID detection only works if that watermark is present. Metadata can be edited.
Can I trust an AI-detector percentage?
Not as a standalone decision. Detectors have false positives and false negatives, and they lag new generators.
What if the photos are of a real person who isn't the advertiser?
That is a stolen-image or impersonation problem as much as an AI problem. Reverse image search and ordinary screening still apply.
Does Velvet22 allow AI photos?
Velvet22's Requirements page prohibits AI-generated, fake, or heavily filtered images. Report listings that appear to break that rule.
Discover verified independent escorts or create your own profile.
