Your AI-generated Art

OnlyOneKenobi said:
Um, yeah, no, thats not going to work… models are trained from images that are curated by humans.
Click to expand...

From the paper:

So far, we have created poison data by pairing prototypical, generated images of A with optimized text prompts of C .
Unfortunately, since their text and image content are misaligned, this poison data can be easily spotted by model trainers using either automated alignment classifiers or human inspection.
Click to expand...
To overcome this, Nightshade takes an additional step to replace the generated images of A with perturbed, natural images of C that bypass poison detection while providing the same poison effect.
This step is inspired by clean-label poisoning for classifiers. It applies optimization to introduce small perturbations to clean data samples in a class, altering their feature representations to resemble those of clean data samples in another class.
Also, the perturbation is kept sufficiently small to evade human inspection.
Click to expand...
We extend the concept of “guided perturbation” to build Nightshade’s poison data.
Given the generated images of A, hereby referred to as “anchor images,” our goal is to build effective poison images that look visually identical to natural images of C.
Click to expand...

8duCRoz.png


 
OnlyOneKenobi said:
Still sounds like a waste of time and effort for little payoff, those who end up with "poisoned" image models will implement something to overcome nightshade or come up with a workaround.
Click to expand...

At the end of the day, if this does get used, it just means that, at worst, models will have to be trained using original images, which will only improve them, since original images are of much higher quality.
I think that's the most commas I've ever used in a sentence.
 
tetrasect said:
At the end of the day, if this does get used, it just means that, at worst, models will have to be trained using original images, which will only improve them, since original images are of much higher quality.
I think that's the most commas I've ever used in a sentence.
Click to expand...
Things are getting more interesting with IP stuff. If this tweet is true things are gonna get wild in 2024. Will be interesting if big corporates also start using nightshade:

https://twitter.com/x/status/1738408490543054949
 
Ghoster79 said:
View attachment 1636511
Click to expand...
I learned some about art as wify is heavily into art and an artist. This is up to page 85 the winner imho, and wify agrees with me. Very abstract novel idea, well executed with contrast and perspectives in balance. Very striking. What is the obsession with monsters and skulls?( question aimed at many other here).
 
Pax said:
I learned some about art as wify is heavily into art and an artist. This is up to page 85 the winner imho, and wify agrees with me. Very abstract novel idea, well executed with contrast and perspectives in balance. Very striking. What is the obsession with monsters and skulls?( question aimed at many other here).
Click to expand...
We are all dying and humans have always been fascinated by death and the afterlife. And as for monsters - while we are afraid of them - humans are the worst monsters. Monsters live in our heads and in the dark - and under our beds.
 
Speaking for myself, I dabble in a few different art mediums, from Photography, photo manipulation, some illustration and CGI.
AI allows me to rapidly visualise concepts and ideas, as well as to visually express and share thoughts and emotions that range from dark and scary to twisted and funny or just plain simple "what if" you put some pop culture character into a ridiculous scenario, just for the fun of it. I've even generated some images inspired by song lyrics, just casually, and they turned out pretty good.
I find that AI technology significantly enhances my creative process. It serves as an invaluable tool for rapidly correcting and enhancing images. What previously might have taken days or even weeks to accomplish can now be achieved in a matter of hours and it can be done on impulse or when casually exploring an idea.
So, while this particular thread might not necessarily serve as a portfolio of "true" art, it enables us to explore, play around and push boundaries and explore new creative landscapes with greater freedom and flexibility.
 
I got this out of ChatGPT, by feeding it an image I had generated previously in Bing/Dall-E3, and told it to produce a similar style picture. It's quite nice, I think. I hadn't previously thought of "showing" an AI a previously generated pic with the goal of doing more variations of it.

IMG_5323.jpg
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X