P: Generated images violate user guidelines

[Hematita]

So as you can see, it's a PG-13 relatively inoffensive image of a woman in a bunny outfit. The top worked fine, and I was able to complete the top ear, which is cool. When I tried to extend the bottom with generative fill, though, I got this warning. They're just a pair of legs wearing stockings, and I wanted to extend it.

It feels like a false flag - though I could be wrong? I find myself thinking it would do the same for women in swimsuits.

Figured I'd share here.

[Encartauk]

I tried generating a woman from scratch using the Firefly web app version to creat a character from a game - and yes, hideous is an apt description. Even when I put 'attractive' or 'beautiful' it didn't get much better. I know looks are subjective, but looking like they were created by Dr. Frankenstein, and then hit by the back of a bus, and then the hospital they were recovering in collapsed in an earthquake... 

Considering the results I've seen people make using other products, where the results were like photos and incredibly good looking women (with usually only the hands an issue occasionally) Adobe has to up their game if they think they are the premiere image creation software.

[c.pfaffenbichler]

Adobe has to up their game if they think they are the premiere image creation software.

Image editing software. 

[Encartauk]

But Adobe shouldn't be using Urban Dictionary to decide it's list of forbidden words, or there'll be no words left. Why can't people create almost (as in, no child-prn or images involving defaming real people) anything they like? You can't use these for commercial purposes anyway. The only problem is the furore the main-stream media might throw at Adobe if their software is used to make certain images - but why can't a company stand up for once and say something like 'we provide software for people to use, what they do with it is up to them' and leave it at that. No one blames a car company if their car gets used to mow people down, a phone company if their phone is used to commit a serious crime, or Windows if their software was on the PC someone used in a hack.

[Encartauk]

Do they not think they're the best image 'creation' software anyway? If not then who is? I'm not saying they are the best for creation, but I'm interested who is - because I've heard of most, and not many else spring to mind - so I'm wondering what I'm missing (not that you're probably allowed to mention it here).

I also said Adobe, not Photoshop - Illustrator is more a 'creation' tool, and I'm talking about Adobe overall, not just PS.

[Graham24508943nobd]

 

 

 

 Here's a few of the women I've created with relative ease. Other times it takes dozens of attempts to get a nice face. Try placing an object in front of a persons hands OR, crop the image so's the hands aren't in view

[Tom S...]

Had the same trouble working with pictures including womens in swimsuits.

Even if I just try to extend the background, without text prompt like legs etc...

[JosephOComm]

Hi, Yes, the Generative Fill takes a very long time to generate now with v25. And when it does finish, it says I violated a guideline which I'm 100% clearly NOT. It does this most of the time.

I have:
PS Beta v25
Win 11 x64 22H2 (build 22621.2134)
i7 12th Gen
1Tb NVME SSD
RTX 3060

Internet:
500mbps DL/UL

Situation:
I just delete some part of the image with a selection tool and did the Gen Fill.

Screenshot confidential, work related.

[Encartauk]

They're not too bad, the first one's the nicest (imo). But you can tell they're not real once zoomed in slightly if there was any doubt. I've only used the Firely web app for people creation, so I'll give PS Beta a go, but the results for Firefly were like settling on the best of the bad bunch and then selecting different facial features and trying other prompts to see if something better turned up. In fairness, looking again now, the result is convincing, but it took maybe hours to get there. 

[daniellei4510]

Even the best results when it comes to creating photo realistic images in any AI application requires running the results through facial restoration software. Even then, the ear would need some additional work or hidden under cloned hair or content aware fill.

[Encartauk]

Your image looks like the teenage version of my 30ish year old one. Maybe it's because it's lower res, but yours loses some of the flaws that give the face a bit of character, so I think the original looks more realistic.

I went from critizing Firely/Adobe quite heavily, to now defending my original image.

Maybe I'm just expecting great results a lot quicker with the user having to do far less to achieve it.

As someone who's been using PS for over 20 years and spending a lot of time learning as much as I could about it, some of the AI images I've seen in the past 6 months make me worried our time as image manipulator/creators is limited. Having these new advancements not work amazingly extends our stay, but for how much longer? When the average Joe can just type in what they want into a free, quick bit of software with perfect, realistic results (coming within the next few years at longest, judging from the speed of advancement so far) why would we be needed and/or paid to do what we do?

[daniellei4510]

"When the average Joe can just type in what they want into a free, quick bit of software with perfect, realistic results (coming within the next few years at longest, judging from the speed of advancement so far) why would we be needed and/or paid to do what we do?"

For the same reason people still hire other people to design their websites, even with the slew of easy-to-use web design applications that are available online. They just don't want to deal with it themselves when they have other things to concern themselves with. Things will all pan out in the end. Models will still find jobs and photographers will still find work. 

[Encartauk]

But, when it's as easy as it seems it's going to be in the future, judging by the speed of advancement already, I don't share your optimism. Have you tried Bing AI Chat's text to image? You can get 4 results within about 10 seconds, and then ask for modifications and it keeps producing. Yes, the results are far from perfect, but the speed at which they're produced is impressive. Once the results are more realistic and convincing, why would the next David Bowie pay some artist to produce the next Diamond Dogs album cover when they could just type it in a prompt and there it is - I mean it's a shame hardly anyone cares about album covers these days as consumers download the digital song anyway and most don't care about the package like I do. 

AI will be great for consumers, not so great for creators in my opinion.

[daniellei4510]

You have some good points but let's face it: it will be a long time coming before we see an AI version of Sports Illustrated's Swimsuit Edition. 🙂

[Graham24508943nobd]

 This took me 5 minutes just now, I simply added the boats, church, waterfall & wildlife, pretty spectacular I'd say

[Encartauk]

Why would they do that? They're already featuring obese women as the epitome of fitness - to represent "real women", so that's the only reason I don't see them featuring 'unreal' women any time soon. But if they weren't heading in the woke direction they could easily fill an issue 1000s of times over with AI generated attractive women. Search youtube for 'AI cosplay' for examples.

I've also heard recently about AI models getting large followings on instagram or similar without people knowing they're AI, and they're making their creators a lot of money - that's at least one area where a creator is benefitting from AI - but give it a while when all people are able to create these pics themselves and why would they be following fake people on instagram?

Just some other examples of AI advancement: a guy on Youtube called Bradley Hall has a video called 'I Made An Entire MEGADETH Album In Just 1 DAY' where he live streamed it's creation and asked his audience for song subject matter and typed those ideas into an AI text generator and it generated lyrics for whole songs, in minutes, if not seconds, in the style of Megadeth/Dave Mustaine. He created the music for it, but AI has also done that too.

There's a youtuber called TmarTn2 with a few videos of a tech demo set in New York where the player can walk up to NPCs and speak through their headset's mic to talk to them, ask questions, etc. and the NPCs respond with AI generated thoughts and can hold full conversations.

You can just ask AI chat to write a poem based off this or that subject. Or ask it to write code for a programme. Or even generate cartoons based off text or a few source materials. There's software that you can enter a short recording of someone's voice then type whatever you want and it generates audio with their voice. I've heard of sites or software with AI generated girlfriends - I don't know to what level of convincing they are, but again, in the not too distant future, put on a VR headset and live in a virtual world with your virtual partner, doing whatever you want and no-one's getting hurt (except the user if they get too carried away - I mean both physically and mentally)

AI is advancing quicker and to an impressive level much more than what Adobe's results are suggesting. That's maybe why I was expecting better from Adobe, but also why I feel pessimistic about our futures and being surplus to requirements in not too long.

[Nour nobe31603624yrxd]

sarkernournobe@gmail.com

[Gregusaurus]

The average Joe will not have the vision, design knowledge, artistic taste, or expertise to manage the creation of something of quality. There is more to design and art than being able to create a cat with a top hat smoking a cigar. AI can be an amazing tool for creativity or for easy production of  tacky schlock.

[Encartauk]

Have you seen how much of a mixed bag 'real' art is out there, whether in adverts, album covers, movie posters, packaging, in galleries. The average Joe doesn't care about it as much as you or I do, but the results I've seen with some AI are convincing, and more than good enough, that they'll more than do for whatever needs someone has. And most of the time, the user can keep generating until they get what looks better at no cost, except a bit more time, to them.

If a client is getting an artist to create an image that's taken many hours or days to create in the first place, and they then want to change things around, but it's going to cost them even more to do so - how many people are going to turn to AI instead where they may get similar results quicker, costs nothing, and can make changes to their heart's content?

For the time being, creators can capitalise on the fact AI isn't perfect and can use it as a USP for why they should be hired still, fix AI hands, touch up the seams, etc. but soon enough those issues will be ironed out. After that, either creators are going to lower their rates so much to still seem appealing and compete with AI, and/or only the most sought after/well known artists will get work from those that can afford it.

[Gregusaurus]

Encartauk, You have good points. "Have you seen how much mixed bag'real' art is out there, whether in advertisements, album covers, movie posters, packaging, or galleries?" Yes, this has been true since the beginning of art. The task of the designer or artist who cares is to create something better. My own experience with AI creation is that it can take me hours to create something I am happy with. When I look at most of the examples of AI "art" they rarely rise to anything that makes me say "Wow". It depends on what the artist or designer is prompting and how they select and choose the elements that best convey their vision. The prompt becomes the magical words of creation.
But caution is advised. If it's now the age of anyone becoming  'The Sorcerer's Apprentice', drowning is a distinct possibility.

[MichaelCluff]

I'm getting this while trying to fill background for a totally G-rated candid photo of a woman with her adult son! 

[joel15A9]

Well from my experience Photoshop beta has policies about the content that can or cannot be edited. I am a boudoir photographer and I suppose this is also true for artistic nudes. If you want to use generative fill, community policy prevents you from doing so. My question is if this makes sense, or even if it's even ethical because apparently adobe has access to my photo while I'm editing it (which even makes some sense as to make the generative fill you need to analyze the photo ). But regardless of whether or not they have access to the photographs as we edit them, what is utterly ridiculous is the fact that I can't use the tools because some enlightened puritanical mind thinks I shouldn't work on photographs of people in their underwear. I sincerely wish other boudoir, artistic nude, fine art and other photographers would add their voice to mine so that Adobe hears that it's not just landscape photographers who need artificial intelligence as a tool.

[RoundRocks]

Who does Adobe think they are, giving a warning about violating guidlines when trying to generate with the prompt "finger" or "tsunami"? There are so many reports about these false possitive warnings, but it doesn't seem to get any better.

[Kevin Stohlmeyer]

@RoundRocks They are a company offering a public beta for users to test on a voluntary basis and report issues to help make a better product.

[RoundRocks]

@Kevin Stohlmeyer I know. Adobe doesn't offer discounts for people helping with beta feedback, so we have a right to vent. Additionally, the venting here is another type of "feedback" for Adobe to pay attention to.

The primary reason people use the beta GF is not to help Adobe - they are using it for work and providing feedback as a bonus. These flags make it very hard to work.

For months, I, along with many others, have uploaded false warnings to the Dropbox they suggested. If Adobe wished, they could make the necessary changes faster.

A word that is searchable in Adobe stock images should not have been trained to get flagged in the first place. If you can search "tsunami" on Adobe stock, and get many results, GF should allow generating them.

[Monty.T.rex]

Hi, I got a message that generated images were removed for violating user guidelines while I was trying to edit the picture (attached, child's face is blurred out). I did not enter any prompts, I was selecting slides on the left and on the right to edit out the dirt marks. Please address this issue, this is a little concerning.