Good news, fellow artists! Nightshade has finally been released by the UChicago team! If you aren't aware of what Nightshade is, it's a tool that helps poison AI datasets so that the model "sees" something different from what an image actually depicts. It's the same team that released Glaze, which helps protect art against style mimicry (aka those finetuned models that try to rip off a specific artist).
As they show in their paper, even a hundred poisoned concepts make a huge difference.
(Reminder that glazing your art is more important than nighshading it, as they mention in their tweets above, so when you're uploading your art, try to glaze it at the very least.)
91K notes
·
View notes
So Netflix is releasing this animated movie but here’s the thing - all backgrounds are done in AI. No actual BG artists were hired or credited for this work.
And here’s the kicker - I gave them too much credit. I realized that not even the AI users are credited apparently.
Check this out:
“+Human”. No names, no nothing.
I know it’s tempting, but please don’t hate-watch this thing. Numbers equal profit even if the people watching are just doing it to mock. Don’t give Netflix and other studios even more incentive to fire more animators and replace them with machines trained on their work and skill. Hilariously, the excuse used is ‘there aren’t enough artists’. That’s not true - there are more than enough artists, but the real issue is that no one wants to pay a living wage. Here’s a short video about the reality of animation in Japan.
The biggest winners in this equation at the end of the day are the AI infrastructure owners - owners of Midjourney, Stable Diffusion, etc. Do not feed this machine, it creates more monopolization of entire industries, raises a powerful few and dehumanizes the rest.
EDIT: a lot of people are unsure what the anime’s called - it’s The Dog and the Boy.
58K notes
·
View notes
i don’t know what else to say except that AI art is no longer simply a source of creativity or a wonder of human creation. it has become actively hostile and destructive toward the very thing it pretends to uplift and celebrate. it is void of any human element, any soul or ounce of emotion or self-expression. continuing to use AI art knowing that it comes from theft and robbing artists of their livelihood is disgusting. they need your support now more than ever. stop giving these thieves your money and admiration.
[Image descriptions courtesy of @cryptid-deity: a series of screenshots from Deb JJ Lee’s twitter, @ jdebbiel.
Image 1 is a tweet that reads, “this... this is fucking sad. I don’t know what to do.” Included with the tweet are two images; one is a screenshot from a private text message conversation, the other is a collage of AI generated art that was sent to Lee in the text conversation.
Image 2 is a screenshot of only the text conversation, which shows multiple messages that were sent to Lee. There is a collage of AI art, then a text message that reads, “Sorry to bother you. It looks like someone on an AI subreddit is making a custom model and it reminds me a lot of your work. He isn’t telling anyone where the images are being sourced from.” Following is another collage of AI art, followed by the message, “Here’s another one.” The last message is a link to the subreddit.
Image 3 is another tweet that reads, “Oh so it *was* me,” with the word ‘was’ between asterisks for emphasis. The tweet includes two screenshots from the subreddit.
Image 4 is one of the subreddit screenshots. It is a post that reads, “Dreambooth model release! Say hello to kurzgesagtish!” Under these lines is a Reddit message that reads, “Sorry, this post was deleted by the person who originally posted it.”
Image 5 is the other subreddit screenshot. This post reads, “Results I got with a custom dreambooth model.” This post has the same deletion message as the former.
Image 6 is another tweet by Lee, which reads, “For comparison, here’s my art vs artist. I’m gonna cry.” Included is a collage of 8 works by Lee, which frame a photo of them. They follow up this tweet with another tweet, which reads, “If I see anyone I know using AI Art, they are dead to me.” The tweet includes multiple comparisons between Lee’s original art and an AI generated art piece based on it, judging by the subjects, colors, and lines.
Images 7, 8, and 9 are the examples from Lee’s tweet.
end description.]
52K notes
·
View notes
...that your audience won't hate.
This is a method I started using when NFTs were on the rise - thieves would have to put actual work into getting rid of the mark - and one that I am now grateful for with the arrival of AI. Why? Because anyone who tries to train an AI on my work will end up with random, disruptive color blobs.
I can't say for sure it'll stop theft entirely, but it WILL make your images annoying for databases to incorporate, and add an extra layer of inconvenience for thieves. So as far as I'm concerned, that's a win/win.
I'll be showing the steps in CSP, but it should all be pretty easy to replicate in Photoshop.
Now: let's use the above image as our new signature file. I set mine to be 2500 x 1000 pixels when I'm just starting out.
Note that your text should not have a lot of anti-aliasing, so using a paint brush to start isn't going to work well with this method. Just use the standard G-Pen if you're doing this by hand, or, just use the text tool and whichever font you prefer.
Once that's done, take your magic wand tool, and select all the black. Here are the magic wand settings I'm using to make the selections:
All selected?
Good.
Now, find a brush with a scattering/tone scraping effect. I use one like this.
You can theoretically use any colors you want for this next part, but I'd recommend pastels as they tend to blend better.
Either way, let's add some color to the text.
Once that's finished,
You're going to want to go to Layer Property, and Border Effect
You'll be given an option of choosing color and thickness. Choose black, and go for at least a 5 in thickness. Adjust per your own preferences.
Now create a layer beneath your sig layer, and merge the sig down onto the blank layer.
This effectively 'locks in' the border effect, which is exactly what we want.
Hooray, you've finished your watermark!
Now let's place that bad boy into your finished piece.
You'll get the best mileage out of a mark if you can place it over a spot that isn't black of white, since you'll get better blending options that way. My preference is for Overlay.
From here, I'll adjust the opacity to around 20-25, depending on the image.
If you don't have a spot to use overlay, however, there's a couple other options. For white, there's Linear Burn, which imho doesn't look as good, but it still works in a pinch.
And for lots of black, you have Linear Light
Either way, you're in business!
EDIT since this has escaped my usual circles, and folks aren't as familiar with my personal usage:
An example of one of my own finished pieces, with watermark, so you can see what I mean about 'relatively unobtrusive'-- I try to at least use them as framing devices, or let them work with the image somehow (or, at the very least, not actively against it).
I know it's a bummer for some people to "ruin" their work with watermarks, which is part of the reason I developed this mark in particular. Its disruption is about as minimal as I can make it while still letting it serve its intended purpose.
There's other methods, too, of course! But this is the one I use, and the one I can speak on. Hope it helps some of you!
52K notes
·
View notes