for anyone wondering about this situation their official explanation on telegram is that they just photobashed some AI backgrounds on accident & said they're going to take a break. Conveniently, they cannot show us any of the backgrounds they "photobashed"
I don’t want my account to become a constant cycle of proving my art isn’t AI-generated. I’ve already made videos about it, and some people still won’t believe me and that’s okay. So I’m going to keep creating what I enjoy
The last proof I’ll offer is a live art stream on Twitch
@aswn00 ? I did not write anything in telegram. I only read some stuff and announced it here since you were too scared to make an official announcement on twitter yourself since it's not a safer echo chamber for you like telegram.
There are spaces for AI prompters. If you want to make AI generated content, join those spaces. Do not invade real artists' spaces AND try to scam people with commissions
@notnotnotdussy0@mengya9918 The video attached is the artist’s “proof” that they posted on telegram in which they’ve coincidentally merged almost all layers and the file was made on the 11th when the art was originally posted on the 9th. There’s also SynthID watermarks found in several of their works
@FragolaBunny@banban131109 Also if we try your logic and test just the background, it doesn't get flagged. This is because the SynthID watermark is laid over the ENTIRE image and cutting certain parts out can make it unrecognizable. the watermark is built to withstand small transformations, not big ones.
@FragolaBunny@banban131109 The SynthID is laid over the entire image, so yes if you cut part of it out the detector may not recognize it. if it was really just backgrounds, why did this one get flagged? are you going to say she photobashed the nearly non-existent background of this one too?
@banban131109@playcentermd this is funny because there was recently an AI artist that got exposed in the WHA fandom who used the exact same excuses about photobashing AI backgrounds and similarly could not produce the original photos when asked https://t.co/WIBNBhsTIs
Dear @xenillxy, who has been accused of using AI to create art. The Knights of Moralis have found significant evidence that your recent “artworks” are AI GENERATED.
They deleted their links on bio so I'm putting it here:
VGEN: https://t.co/7CAHf1hDG1
More evidence below 👇🏻
@FragolaBunny If you have any proof that it does have a higher fpr than whats been documented by deepmind research then id be fine with that. But i already knew that its false negative rate was higher, afaik it kinda has to be for them to meet that target of 0.1% fpr /gen
@FragolaBunny@aswn00 Basically
No SynthID detected = odds are that it could still be AI generated
SynthID detected = very high odds that its AI generated
there's a reason why i said the fpr was 0.1% instead of saying it has a 99.9% accuracy, because these two things are separate.
@Alpacat2 Though ofc it would be a stronger case if it wasnt just deepmind's study but i struggled to find much else that would conduct a study on it in a similar way (and specifically on the image generation model and not text generation). Which article had the 2-8%? /gen
@Alpacat2 Its reportedly 0.1% from OpenAI's own research paper on it https://t.co/Q21FPW18Hp Also they clearly arent just using AI to refine details the AI is doing most of the heavy lifting, otherwise they wouldnt be as hesitant to show any wips or layers at all