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.

[LSDMT]

fix your garbage AI implementation, Adobe. name one freaking thing in this picture that violated ANYTHING. want to fix a little bit of hair in the corner - 'ThE gEnErAtEd ImAgEs WeRe ReMoVeD bEcAuSe ThEy ViOlAtE uSeR gUiDeLiNeS.'

and it happens ALL THE TIME. there's no nudity, there's no gore, there's no nothing but ineptitude on Adobe's part to fix something that's been REPEATELY brought up by users as a constant FAIL on Photohop's Generative AI 'feature'.

[rayek.elfin]

Note: you are not talking to Adobe here - we're all users like you. 🙂

I understand why Adobe's geenerative AI implementation is so strict/restrictive, because as a company they'd be opening themselves up to all sorts of litigation if the results would include even a smidgen of "sensitive" imagery.

So why not use one of the open source free solutions out there with no such limitations? And those will run on your local computer, so no need to rely on an external server. I do that myself, and have no issues with any restrictions. And arguably the generative AI models out there have improved so much that they now generate better looking imagery than Adobe's AI.

Do a quick Duckduckgo search to find out more. It takes a little more time to set up, but you'd gain access to generative AI imagery without the constant warnings.

[daniellei4510]

It's a bug and it has been well-documented here that entering a "." as a prompt will usually resolve the issue. But I'd rely on more traditional tools such as the Clone tool and perhaps the Content Aware tool. Even when officially released, Generative Fill will not be a cure-all for editing. 

[LSDMT]

i use Stable Diffusion almost entirely. that image was created by Stable Diffusion. but the inpaint options usually don't work as good to fix minor things as Photoshops. that in no way lets Adobe off the hook. if they want to nerf the AI into the ground, then why even implement it? it's just so they can say they've also tacked that on to their pre-existing apps in an effort to remain relevant. if they want to actually make it useful - then they have a LOOOOONG way to go because the super smart AI that generatively fills images in Photoshop is pretty freaking dumb and unable to distinguish much of anything. i even have it fail repeatedly on SCENERY pictures. it's... not really that good.

if you have another solution that isn't inpainting via SD - i'm all ears.

[lucidframe]

I gotchya'. Thank you!

[defaultyfp7chz9kt23]

With respect, we are not arguing whther or not generative fill will replace "traditional tools" were arguing that adobe's arbritary moral landscape is getting in the way of Creativity.

[rayek.elfin]

I use AI-Runner nowadays to avoid having to install and set up all the dependencies myself. Just run it and choose a Generative Model. Runs on Windows and Linux (Ubuntu compatible) from its folder. No Mac, unfortunately, though.

https://capsizegames.itch.io/ai-runner

This desktop application runs on your local machine rather than a server. No internet required, which I prefer. And I have far more control over the results this way. But it does require a beefy graphics card, of course.

AI-Runner's GUI is simple to operate, and all functionality is nicely organized in tabs: txt2img, img2img, inpaint/outpaint, depth2img, pix2pix, superresolution ai scaling, themes/styles, LoRa, etc.... And most importantly: no limits on output.

AI-Runner also loads other models (for example ones downloaded from https://civitai.com/), so it is relatively simple to find something a bit more original/different looking.

[daniellei4510]

But the guideline violation message is just a message, and a poorly drafted one at that under the circumstances. If it said something like, "Sorry, but the bot is having problems trying to determine the intent of your action or prompt. Please try again." Or whatever...then accusations of censorship or arguments against Adobe's moral landscape would be close to none and everyone would be getting frustrated for a whole different set of reasons. 🙂 Give it time. Yes...some of the issues may be directly related to censorship for now, but that seems understandable this early in the game. Even Midjourney has finally gotten around to adding an "Appeal" button that sends a questionable prompt to a "smarter" bot that can sort out the actual intent of the prompt in question. The majority of those appeals are approved.

[Jason Toemmes]

[Jason Toemmes]

It's been awhile since I posted on this violate guidlines error. Strange my orginal post has morphed in with everyone else problem. In general I see many are just as annoyed as I was over the idea being censored. I have been keeping up with the latest updates and overall when I ask the genernative fill to remove a object or fill in a empty space on my canvas it does a pretty good job 80% of the time. I really haven't had the time to sit down and push it's limits till last week.  I'm wrapping up this project without PS Beta, but I wanted to see how OR if it the AI bot would violate what I suspected would be a problem. One of my accounts makes high end wood and metal old model ships, WWI model aircraft, and artillary from the US civil war. The project invloved creating a print ad for a model of the 1850's Napolean Cannon that was used in the US civil war. I have uploaded the art without any of the overlying ad copy that would hide roughly 30% of the art. What I wanted Generative fill to do was add a war torn rebal flag in opposition to the Union flag flying over the canon in the background. Surprise, surprise, that request was violated. So I tried saying add a Confederate flag and that too was violated. So basically it's really cut and dry that I am going to be forced to use none Adobe software for some of my professional work on this account. This is just a prime example of what I was concerned about, of corprate nanny culture interfering with professionals like myself in doing their job. Like it or not history around the world wasn't pretty. Context for images created are very important, as was in this case the intent of this ad was to be educational and thought provoking. Most people including myself had no clue a French cannon was used on both sides of this important peice of US history.  In the end the confederate flag would mostly been hidden with text anyway. This censorship issue remains to be a problem. 

[ronaldr97793654]

AI is saying I'm violating user guidelines due to nudity

V 25.0

MacOS 12.6.8

[LSDMT]

that's just Stable Diffusuon in different clothing 

[J-D-F]

"it seems understandable" is as subjective as how and what to include in an overly-zealous censorship engine. "What" is understandable? Without answering the "what" it's all a matter of personal preference. Adobe isn't breaking any new ground here except with their legal wrangling.

[rayek.elfin]

I am aware of that: I posted it because you stated that you are frustrated with Adobe's implementation and its imposed limitations of what can be rendered, and running your own SD avoids those.

[Joe23317752iugo]

I will literally select an area of walking path, enter no prompt, and it'll tell me it violates guidelines.  Been trying to remove a trash can from a picture for like 1 hr =(

[daniellei4510]

Did you enter a "." for a prompt? This often fixes the issue.

[Gregusaurus]

Consider using the 'remove tool' for removing trash cans. 

[daniellei4510]

"It seems understandable" in the sense that every legitimate AI program today has restrictions and banned words. In any case, if editing the arm of a couch results in a guidelines violation, and editing skin does the same, there's no way to know if the message has something to do with censorship. Will we ever be able to generate a nude from scratch using generative fill? Nope. Never. There are other avenues for that. Will we be able to edit skin or the arm of a couch in the future? Yes. In time.

[J-D-F]

Chair arms and skin are two cherry-picked scenarios. In the world of photography there is a massive breadth of content that includes people in various states of dress, situations and poses. Choose some scenarios like how short can shorts be or can you show a diamond necklace above a plunging neckline? Adobe has become the principal of a grammar school, and reasonable is in their hands. "The other Ai companies are doing it." Is not an argument. It's an excuse. 

[daniellei4510]

Truce. Because this is like arguing politics with someone in a bar. Unless we decide to run for office or get hired by Adobe as a programmer, there isn't much we can do to change what is. 🙂

[Taymuraz245358520iti]

I also agree with that

[GRPImagery]

its a bit of a joke what gets flagged IMO.   Fairly random with no way of notifying/repealing.   Makes the funtion randomly useless.

[RLeintzPhotos]

I think the biggest problem right now is we don't know what's a bug and what is actually a violation. For example, if I have a field of flowers with a path on it and select it, gen fill with no prompt, and it gives me a user violation, ok, got it, it's a bug. But if I select skin that's close to the top of a breast line and ask it to remove wrinkles and I get a violation error?... What? Is the Ai tool just bugging out again or is this some kind of morality filter? I'm hoping we'll get more clarity when it gets out of Beta. Then at least we can know what we're debating 🙂 

[ArnelGonce]

Yes, but Adobe is, well, Adobe. Much bigger, more incoming funds, more
support staff, more expensive, ie, more expectations from the public. You
wouldn’t expect the same things from a Lexus as you would a rolls Royce.
That “button” should already exist.
In the last couple of days, I’ve been advised to update PS beta to get the
program working in sync with LRC, then advised to go back to the previous
version when there were new issues that came up, then back to previous
issues that were worse than the first time, and then back to the upgrade,
still with issues but at least workable if not efficient. I get hiccups,
but leave a workaround that works until a corrective update comes along.
Lately, it’s been very bad.
--
Arnel Gonce

[Sarah27914909k9u3]

If Adobe intended that as a policy, they could easily say so, but they haven't. Both words AND deeds show what a person (or a company) really is. Adobe, please clarify your policy for post-beta!