Is it unreasonable to ask for AI/enhanced pictures to be identified as such?

[Bill McNeal]

Originally Posted by jayteee

I am all for well-executed AI enhanced material. But that is the key, *well-executed.* An AI enhancement tool is a tool just like any other - results depend heavily on the source images/video as well as the skill and judgement of the tool operator. From what I've seen, many people working with AI tools have very poor judgement as to what looks good. I see material posted all over the internet where faces are so altered that the subject would be unrecognizable if not for the context of the rest of the image. I've seen video where everything looks like plastic. That is in part because of the way the current AI tools work and partly because of the judgement of some of the tool operators.

I see this as a problem for generations going forward. For instance, I've seen images of Petra Verkaik where her face looks nothing like +/her face/+. People searching for her material in the future will be wondering what she really looked like. Is that her face? Or that one? It all depends on which images survive the ages going forward - which images are promoted or posted all over the place, etc.

I completely agree that AI enhanced material should be labled as such.

Sure, it in no way guarantees that material will be labled as such when posted to other places, but at least it could be referenced here for what it is for as long as VEF exists.

I've posted both images and video I've personally run through various AI tools and I'm proud to declare what I used to produce those images/videos.

I think it should be requried if you're the original poster of the material.

It's a gray area for material reposted from other sources. But still, labeling it should be encouraged, if possible.

My two cents.

Petra has really been given poor treatment recently in her thread. I reported a few egregiously bad images posted in June. Despite being unquestionably of poor quality, I was asked to locate originals for each of them and also see if they were in the gigantic thread somewhere. Sorry, as much as i like Petra, I can't scan through her entire thread every time a new bad AI pic is posted.

And I wasn't even asking for the pics to be removed, just for them to be tagged as AI enhanced. Since someone else just mentioned Petra specifically, it's clear it's becoming a problem in her thread.

[jayteee]

Originally Posted by Bill McNeal

Since someone else just mentioned Petra specifically, it's clear it's becoming a problem in her thread.

I don't know if it's becoming a problem, but some AI upscales are more successful than others. Yesterday, I was in a crap mood and saw a new AI upscale where her face just wasn't her face and replied as such in the thread.

After the fact, having gone back to an original version of that exact image, that particular image of her has a unique quality to her expression. And even though the upscale was off, it wasn't as far off as I had originally thought.

I think some of the issue is that some of the operators of the current AI upscaling software don't understand that if the image of the original face is really poor, and depending on your settings, the AI will simply attempt what I like to refer to as a Mr Potatohead operation... it replaces the features with features from a parts bin that the AI thinks fits. The problem is that this usually results in a completely different face than the one you want.

Until such time as the AI upscaling software can be trained on specific people and the operator can specify to use only those specific training sets when upscaling, this will be an ongoing problem. And it may need to get even more specific than that when you have a model like Petra who's work spans decades.

Although it seems like I'm for censoring the less successful AI upscales, I'm more for labeling them as such when posted. My concern is that unfamilliar, unsuspecting people don't realize what they are looking at when they see these images.