A bit of a TLDR for some questions I saw in the notes:
The team that created Glaze is from the University of Chicago. Their names are each listed in full on the Glaze download website. (This group of students/professors did this for their SPRING BREAK 😱 so go give them some love lol)
It is free to download. No, they won’t ask for or raise money from/for this project.(stated by one of the lead professors of the project).
Glaze is designed to protect artists’ STYLE--which a bunch of ai people have been deliberately fine-tuning their models to mimic (and specifically of current living artists--small or big).
It currently does not protect against composition/trace-like theft (as seen when run through img-to-img) but that would be protected by copyright anyway while STYLE is not.
The University Team has stated that they are dedicated to continuing to improve the tool, like fixing bugs (like overheating older computers by taking up lots of energy when Glazing--it currently runs on CPU so they’re trying to change that to GPU, I believe) and expanding the type of protection given to artists (like working against img-to-img theft).
It currently only works directly on your computer (phones not advised due to current overheating issue, no tablets, or iPads, and no website runthrough since that would be insecure to breaches/scraping/hacks)
It currently works best on painterly artwork, but can still be used on other forms (team is working on improving this)
IT WORKS BY calculating the changes each image needs for the best protection against style theft by AI, and adds tiny changes throughout the piece, so that your style will, for example, confuse the ai into seeing van gogh. But the ai thieves will see a regular image in your style, feeding it into their model labeled as your work (thus starting the “data poisoning”).
Do not post the original unGlazed piece of your artwork after posting your Glazed version (obviously)
The Team worked directly with over 1,000 artists that were being impacted by the ai theft. Because the team listened to those artists, Glaze accounts for regular art thieves too (i.e. Glaze can’t be removed/cropped etc. like signatures or watermarks when reposted. It’s just part of the image, so even if it ends up on another site and scraped, the Glazing is still in effect)
When you run your artwork through Glaze, no information is sent back to the Team. (Aka, no scraping on their part. The app receives information from the Team (like updates) but no information from you is given to them through the app. Basically Team servers ---> You and NOT Team servers <--->You) One-way data street.
Brief misunderstanding happened over an open-source license for the front-end part of the app. (Used open-source coding for front-end, not knowing that code’s use-license states it is only for other open-source uses, not closed-source (the back-end code of the app is private to prevent counter-counter measure developments)). The Team took down the app until they replaced the front-end code with code written from scratch by the team. They are now not in violation of that open-source license since they are no longer using it. (you have 30 days to remedy a license breach once informed; they did so in 2)
The Team is currently in touch with Japanese artists to better expand the tool for use to protect their art styles
From what I understand of it, Glaze is an AI tool designed to be anti-AI (Think Arnold Schwarzenegger in Terminator 2: one Terminator robot vs. all the other Terminators 😂)
You can download it from their website and also contact them through email there with any questions, problems, or bugs. The website: https://glaze.cs.uchicago.edu/
ALT
