💥Mostly ocs and cms work. yap acc: @oookyruJ📩 🚷reuploads/reproduction / AI training is prohibited ❌
17🇻🇳
- Tweets 8,328
- Following 968
- Followers 13,635
- Likes 85,133
Juoshi✨ retweeted
~Special Halloween Adopt~🥀✨💍
Sp: $120
Claim via Vgen or DMs!
vgen.co/_Ashlie_Chan_
#adoptable
2
18
80
1,783
ok so after my previous qrt and now this one, i think it's time we talk about what synthid is and how it works, because i don't believe mikqam's explanation holds up
first of all, i'm not saying they can't draw. there's older posts of theirs where i do not believe they used ai whatsoever. what i'm saying is they've been using ai in their recent work. that's the issue. but please don't harass them over this. look at the evidence and decide for yourself. this is meant to be a warning to potential clients, as honestly i was almost one, and NOT an excuse for harassment
source links will be included in the replies
WHAT IS SYNTHID?
synthid is an invisible (to humans) watermark made by google deepmind. it's a specific pattern of teeny tiny changes to the pixels of an image, but the changes are too small for us to notice. it gets added when an ai tool that uses synthid (like gemini or chatgpt) makes or edits an image. google's detector (synthid.com) can recognize this pattern
it's not perfect, since heavy editing, filters, or compression can sometimes weaken it, but it can survive as well. if it's weakened so much the detector misses it, ai images can slip through, but it almost never works the other way around. most other "ai detectors" are guessing whether or not something looks ai, but synthid isn't guessing. it's looking for a specific pattern which pretty much doesn't show up in images that never went through one of the tools. missing ai is possible, but flagging something that never touched ai is really hard to do
WHAT I FOUND
i already went through this in my previous qrt (linked below), but basically i ran a bunch of mikqam's recent adoptables, saved straight from their own posts, through google's official detector at synthid.com and 5 of them came back "SynthID detected". one false positive would already be really unlikely, and five from the same artist is way more unlikely
the ones that didn't came back negative, but that doesn't confirm or deny that they're solely human-made OR had ai assistance. it just means no synthid watermark was found. synthid only catches images from tools that use it, so a mix of results doesn't mean it's "unreliable" or giving fake positives. if anything, the unreliability is that it won't always catch synthid, especially if it's weakened. all a positive result shows is that the specific image went through one of those ai tools at some point. that could mean it was generated or edited with ai, but either way it's something clients should know about
and no, x's compression cannot cause a positive synthid. compression can weaken the watermark, but it cannot create one
some of the pieces also have things that look like ai artifacts to me and others, but that's just our opinions. you should take a look and decide for yourself
ABOUT THE BRUSHES
mikqam says they have not used gen ai, but that they have used brushes by cozy brushery, which may have had ai elements in them that probably caused the results. while cozy brushery has been widely accused of reselling stolen brushes and using ai-generated preview images, i don't believe their brushes would explain a positive synthid in the way mikqam appears to have used them, even if the brush textures themselves were originally generated with an ai that uses synthid
when you turn an image into a procreate brush, it gets converted to a greyscale mask, shrunk down, and stamped over and over at different sizes and angles, and sometimes you may even paint on top. this scrambles the exact faint pattern synthid needs to be recognized. on top of that, the synthid watermark is applied to the whole image, so unless it's a big stamp brush placed pretty much as-is, it's really unlikely a procreate brush would carry enough of the watermark to set it off. mikqam's brushes they listed, as well as their adoptables themselves, did not appear to use these big stamping brushes, but rather smaller, repeating ones like chains
google hasn't tested brushes specifically, so i won't say it's literally impossible, but i haven't found a case of normal brushes (not large stamping brushes) setting off synthid, and it's not a believable explanation for multiple pieces testing positive
someone has also been testing cozy brushery brushes themselves, putting strokes on a canvas and running them through synthid.com, and as of writing this, none of them have come back positive (linked in the replies)
ABOUT THE LIVESTREAM
mikqam said they'll be appearing on a livestream to draw, which i believe will go smoothly and prove that they indeed can draw. but that wasn't the problem. a stream doesn't explain why specific pieces have an ai watermark in them
CLEARING IT UP
if these were fully drawn by hand, there should be more proof, right? it sucks that we as artists have to hold onto proof of our hand-drawn work, but more and more people are trying to pass off ai as their own art. most artists i know have no problem showing their layers, sketches, etc
as for the brushes specifically, there's a really easy way to test it that only takes a few minutes and doesn't need any old files or recordings. just stamp the brushes on a blank canvas, export it, and run it through synthid.com. if the brushes are causing the positives, that would show it
once again, please please don't harass anyone over this. just look at the evidence and decide for yourself. sources will be in the replies, and i will do my best to answer any more questions about synthid in general
Regarding my artwork being flagged as AI
Lets have a conversation. I'm open to genuine questions and concerns, but any hostile replies will be blocked without a response. Please keep things respectful.
First of all, it's genuinely upsetting how quickly people are willing to accuse an artist of using AI without knowing anything about their process or how much effort they put into their work.
After looking into why my artwork might have been flagged, a friend helped me discover that some of the brushes I've been using from Cozy Brushery were made using AI-generated elements. I had no idea about this when I purchased them, and I believe this might have contributed to the detection results.
However, my artwork is still drawn by me. The compositions, character designs, linework, and illustrations are all my own work. Unknowingly using brushes that contain AI-generated elements doesn't suddenly erase the hours I've spent drawing.
I've also seen people asking for speedpaints as proof, and while I understand why, I unfortunately can't provide them for every artwork that gets flagged.
For context, I'm using a 4-year-old, 10th-generation iPad with only 64 GB of storage. A huge portion of that storage is already taken up by the system and its updates, leaving me with very little space for my drawing apps, artwork, and other files. I haven't been able to update my iPad in a long time because of how limited my storage is, and because of that, I regularly deal with apps crashing, freezing, and lagging while I'm drawing.
I already have trouble keeping my drawing apps running properly, so saving speedpaints for every single illustration just adds another thing I have to worry about. I often have to prioritize keeping my actual artwork and project files over recordings, especially when I'm constantly running out of space.
What bothers me is how easily people are willing to believe an AI detector over an actual artist, especially when these detectors aren't always accurate. Not having a speedpaint available shouldn't automatically mean someone is hiding something either.
I understand how important it is to hold people that use AI accountable especially for art because it's a way to protect the craft and hardworking artists. However, I think we should also be mindful of how damaging passionate artists' drive and dignity with these rash alarming accusations is also a way of letting AI strip real people of their jobs.
I don't mind people asking questions or pointing out things they find suspicious. I understand being cautious about AI art, but there's a difference between asking for clarification and immediately accusing someone.
I'll be more careful about the resources I use from now on, but I also hope people can be a little more considerate before making assumptions about someone's work.
I've spent years learning how to draw, and it's honestly exhausting having to constantly defend my own art because of a detection result.
If my explanation still isn't enough, I'm willing to go through the supposed "inconsistencies" in my artwork one by one. If there's something specific that looks off to you, you can ask me about it and I'll do my best to explain.
Another thing I want to address is the tracing accusations. Apparently, even when I do provide speedpaints, people are now accusing me of tracing because the program I use has a private layer feature that allows certain layers to be hidden from recordings.
It's honestly frustrating because it feels like every time I try to clear something up, there's another reason to doubt me. First it was AI, then it was the lack of speedpaints, and now even the speedpaints themselves aren't enough.
I understand wanting proof, but I genuinely don't know what else people expect me to do at this point.
With that being said, I'll be appearing on izakiko's stream at 9:30 PM (UTC+8), where you can watch me draw live. If you're still doubtful about my process, you're welcome to come watch.
10
14
132
10,861
i'm making a qrt because i have a fair bit of stuff to include, there's a lot of traces of AI artifacting, and there's also what i believe to be outlining on top of the AI gen artwork as well as synthID watermark in some images
i'll be including more in the thread below
6
21
170
25,966
This persons previous works have many ai traces...
Pearl Narwhal Adopt !
SP1: 320$ + all rights
SP2: 380$ + SP1 + Rendered Bust Up
#adoptable 🔃 GREATLY appreciated!
21
49
737
66,853
