Another idea I had with this concept is to make an LLM-proof captcha. Maybe humans can detect the characters in the 'motion' itself, which could be unique to us?
- The captcha would be generated like this on a headless browser, and recorded as a video, which is then served to the user.
- We can make the background also move in random directions, to prevent just detecting which pixels are changing and drawing an outline.
- I tried also having the text itself move (bounce like the DVD logo). Somehow makes it even more readable.
I definitely know nothing about how LLMs interpret video, or optics, so please let me know if this is dumb.
No comments yet
dylan604 · 22m ago
Has anyone tried a long exposure to see if the motion smears into something discernible? Obviously harder to expose a bright screen without some ND since the shutter speed is the phone's main exposure control
Firefox on Android seems to just be a static image, I can't see any text.
kemayo · 27m ago
This makes me feel motion-sick, which is kind of impressive because I'm normally not easily susceptible to that.
dylan604 · 20m ago
My eyes went straight into seeing 3D image mode. It's the easiest one I've seen yet! /s
UltraSane · 5m ago
Seems trivial to diff multiple screenshots to identify what parts move. Or just use a compression algorithm to do the same.
alliancedamages · 19m ago
You can also break it by recording the screen, of course.
altcognito · 44m ago
Fun side effect: staring at the letters for a bit makes the rest of the image move.
bix6 · 1h ago
Ha cool! How’s it work?
Lalabadie · 39m ago
The only way to see the text is in the movement. The pattern across any single frame is entirely random noise.
cryptoz · 1h ago
Had a lot of fun trying to break this. Turns out you can screenshot real easily by zooming out. Maybe there are other ways but I stopped trying :)
vunderba · 13m ago
yeah - I actually was initially confused since I wasn't having any issues screenshotting it but had forgotten that I have the default site zoom set to ~65%.
sans_souse · 44m ago
Not sure what you mean - I can screenshot it freely that's not the point the point is if you look then at the screenshot you cant discern the text because its a single frame now
I zoomed out to 90% and could make out something was there but wasn't easy to read. Zooming out further went back to just being noise. I also tried zooming in but with no success. What zoom level did you use and I guess we have to ask the standard what browser/version/OS/etc?? My FFv142 on macOS never took a screen grab like you did
dwg · 40m ago
Zooming out before taking screenshot and the text is no longer obfuscated. I tried and confirmed it works. In fact, the text is perhaps even more readable than the original.
anigbrowl · 29m ago
It depends how fast or slow your GPU is. I tried it and saw the effect you described, but within a second or two it started moving and was obscured again. Obviously you could automate the problem away.
dylan604 · 14m ago
Mine freezes the animation on zoom change. Not sure you could automate against that
They even provide the source code for the effect:
https://github.com/brantagames/noise-shader
- The captcha would be generated like this on a headless browser, and recorded as a video, which is then served to the user.
- We can make the background also move in random directions, to prevent just detecting which pixels are changing and drawing an outline.
- I tried also having the text itself move (bounce like the DVD logo). Somehow makes it even more readable.
I definitely know nothing about how LLMs interpret video, or optics, so please let me know if this is dumb.
No comments yet