AI Anonymizer – use virtual faces to secure your identity(generated.photos) |
AI Anonymizer – use virtual faces to secure your identity(generated.photos) |
No thank you. I would rather use https://thispersondoesnotexist.com/ instead. I do not want to expose myself to protect myself. Please think about it, it is absurd. You do not protect your privacy by first giving up privacy by trusting total strangers with your photo. Probably could use https://generated.photos/faces as well.
Such an anonymizer. You probably use your IP, you upload a photo of yourself to someone's server, and God knows what happens to all this. Just to... anonymize yourself. First step to anonymize yourself: do not upload a photo of yourself to random websites.
However, I quickly came to the realization that the metrics I'd try to alter would probably not be sufficient and reveal enough information that the whole thing would be rendered moot.
A nice first step to remediate trust (or at least shift trust relationship to other parties) would be for vendors to implement highly restricted sandbox models (i.e. not allowing _any_ network access and restrict/alter ways of in- and output, like adding noise and artefacts to image data that can leave the app e.g. via share/save to camera roll).
I don’t think it’s the best way to hide. Generated photos are detectable after all, and using a generated photo raises a flag.
But I say it’s good enough for most cases of everyday life:
* Uber drivers chasing attractive customers online
* Jealous ex
* Airbnb and coach surfing
* Classifieds encounters to sell a bike, etc.
You can use it for free, just attribute it.
$2.99 gives you a double resolution.
We don't use these photos to train.
In particular, I have naturally blond (very blond) hair, blue-green eyes, but a black beard.
This seems to confuse the AI. None of the pictures it generates look even remotely like me, and I even got a broken one (long-haired ones, where the hair fizzles out into AI artifacts).
Now, just to figure out how I can use my AI-breaking power face to mess with facial recognition technology? ;-)
I’ve discovered that in any unclear situation it shows toddlers.
In particular, it happens when the camera distorts the perspective, such as too close or shot with an unusual angle. Precisely, selfies are often fish eyed.
I wonder if a plain, well lit portrait changes it. It must not be essentially a passport photo, but a passport photo would work too.
Of course, that is something to think about in the future releases. So sad, Anonymizer is an experiment made for fun. I wonder if it has a product/market fit. Thoughts?
There was an improvised AMA in another thread: https://news.ycombinator.com/item?id=25208704
I'd be glad to answer the questions here as well.
How is it done? The images are pregenerated: 2.6M to be precise, all of them are here: https://generated.photos/faces
Technical challenges of better resemblance
We could either pre-generate (see the first comment - it’s an accurate guess), or generate just in time.
The first is way easier - same architecture, more storage (we are moving from instance’s storage to S3 so it will be an easy step).
On-demand generation would be much more expensive as you have to have a GPU server ready, and queue requests. We have it with https://icons8.com/upscaler ; it’s a bit of work and expense.
Additionally, you have to make it a paid service. Any ideas who could pay for that? Our usual customers are https://generated.photos/use-cases
Brad Pitt is not blonde
It definitely has some fitting issues; Brad Pitt is a good example as he dyes his hair. I've googled his hair situation so you don't have to. He most probably has gray hair, dyed dark, then highlighted.
That's a challenge huh! As a result, you can feel some of these guys manipulated hair color: https://i.imgur.com/bC6AxXO.jpg
Anyway, I had better luck with Angelina Jolie. Hairwise, her hair looks also toned, but our dataset has way more colored women: https://i.imgur.com/WmTzRAZ.jpg
Spam bots have a new face
On the boring side, it’s probably not worth it. Generated photography is detectable. A trained eye could spot it; big social media have models trained to a lot it.
Scraping a real life dataset makes more sense to me. Moreover, there are tons of ready ones.
Instead, here are are some of the other uses invented by our customers https://generated.photos/use-cases