Option to add a watermark to images

[quasiblob]

Would be useful to be able to add a watermark (like your name/tag).

For example:

• Use a texture with alpha channel
• Or a single color black and white texture with opacity control
• Then a value to offset the tag image by pixel x/y amount from a specified corner (bottom right as default?)

This could be a feature in extras tab or something like that, preferably only usable on single images/batch of image when needed.

[ZCaliber]

I feel so conflicted about this. While yes, a decent amount of work can go into prompt manipulation and tweaking dials... But to slap your name on the final product? I dunno.

I think the best analogy I can come up with is that we are all creating recipes for a cook. We are creating a method for creating an image that, hypothetically, can be recreated exactly by someone who can duplicate the 'recipe'.

Eh, a problem for future us to figure out... Or more accurately a geriatric judge who can't look up google on an android phone....

[Heathen]

Being anti-copyright, I'm so not into this....

But, here's an alternative idea;

Being able to set automated captions in the image, such as prompt, seed, steps, sampler, etc.
And a custom text for those so inclined on adding their name or something else.

As for adding watermark, there are plenty of programs that have batch processing capabilities that do that, Photoshop included.

[CCpt5]

It's very easy to find applications that will add a watermark to an image. Putting a "feature" on the main GUI offering to put a watermark will encourage (or at least make every user consider) whether or not to slap some text on an image.

Personally feel like this is something the small percentage of people who would want this capability can figure out how to do on their own. Adding the option will just make it a necessary choice / lead to more tarnished images.

Face it, people hate watermarks.

[quasiblob]

@CCpt5 "Face it, people hate watermarks." I simply voiced my opinion - as I find it useful, IMHO you can stay in your collective "people hate watermarks" group, unless you have something constructive to say. You clearly didn't read what I wrote, neither do you know what I need it for. You clearly know the percentage of people who require such feature and you clearly know what those people would do with such feature.

[Heathen]

Do you think everyone shares your opinion, that they would rather do batch processing in some other software? Do you think that everyone has access to Photoshop?

But I guess it would be so dangerous to have an option to overlay some images... so narrow minded.

If you felt attacked by my reply, sorry, wasn't my intention. I meant to give alternatives, they're not hard to find. Gimp has a plugin for it as well. There is also uMark and the free version is very reasonable.

[CCpt5]

@CCpt5 "Face it, people hate watermarks." I

I stand by this statement. Your intent/need regardless. My point here (in a discussion forum), is that having that option front and center on a GUI will lead to people using it unnecessarily.

No reason to take offense.

[quasiblob]

@CCpt5 Same could be said about inpainting or any other feature... "muh fear, someone is going to paste a wrong face in a nude photo" - so lets not lead people using such feature, lets not add a tab/button for it. Whatever.

[RoyalShooter]

I want to put my views on this topic.

Some background :

Recently NovalAI is trending and everyone making their waifu and posting it on art sites like Pixiv and DeviantArt. This sparked make the artist to fight against AI art with the form of saying it out loud and also some vtuber banned the use of their art tag with AI art. Most of the controversy comes from

1. Novel AI train model based on pixiv public art but without artist's consent.
2. People fake/mix AI art posts with real art drawn by artists. (Somehow the mainstream idea is now AI art should be separate and mark)

there are many more controversies but these 2 get the most vocal and I would like to point them out.

I don't think there is much to do to fix no.1, it is not really possible to ask the dev to get consent to thousand of pictures. But for the no.2 problem, I think a watermark will help in this situation. That said tho putting a big fking “generated by Noval AI” visible on the pic is obviously ugly as hell.

So I propose to:

1. add a custom button to do "add a watermark/border to this picture", you could use this button after you find a gen that you like.
2. add code to read a transparent PNG from folder, overlay the png on top of the generated picture. This meant you can replace the png and make it however you liked. You can add or own name or just do some little changes to indicate to people that yes this is a AI art.
3. add code in setting tab to change png quickly.

Some more thoughts on why this is needed and necessary:

The current perception in the anime community is mostly negative, with some expectations of people who are really into tech or AI or those who don't care about copyright issues. This ended up with many community starting to limit the posting of AI art or outright banning it. And I want to do something to stop this going. I believe AI art has its function and should not be outright banning it. That said with it controversies it is hard to convince people to support it. So putting rules to let both side settle is the best option now. Starting with this Watermark/Border function.

Why make it a button option but not a forced locked future if people are heating it so much and want all AI art to be marked:

Because 1: people hate watermarks, 2: people hate to be forced, 3: there will always be a way to unlock it no matter how we want or not, so it just does not make sense to make it a forced locked future. (just like piracy) 4: There will be cases watermarks that need to be removed in order to do post process or other usages.
So an honour system should do the job. If you want to put a watermark on your AI art, here is a quick and easy option for you.

[Heathen]

Let me tell you from my perspective as an artist and also having seen the gazillion waifus that are popping up from NAI. The majority of them comes from NAI itself on imageboards. So putting an watermark in this repository, even if it is the best one out there, won't really do much. On that note, NAI has a very static style, you can tell something was made by NAI at a glance and I'm sure people will get bored of it fast, I know I already am. Also, some VTuber banning their tag is as stupid and unrealistic as it sounds. Public personas are exactly that, public. It's like Nintendo wanting to crack down on mature art. It's unrealistic, dumb and only pisses off fans except for the cultist ones.

Here's what I think, it's the job of the image boards to enforce people tagging AI art as AI and giving an option to not display AI art. You don't like AI art, there, turn it off and you wont see it. Problem solved. If wanting a watermark is something you want, there are plenty of free programs that do it. Besides, putting a NovelAI watermark in this repository while using the leaked NAI model instead of the official site, which doesn't do the watermark, is just... I have no words. Are you going to make watermarks for each model you didn't train as well? Whose responsibility is that, the people that made the models? Do they want to do that?

What I can see as being useful and it is super easy to do on python is adding on the image any of the generation options you want to add, like prompt, seed, steps, sampler, etc. Even your name if you're so inclined. That's useful to see some info at a glance without using PNG info.

The other problem I see with adding a built in watermark is that it gives the people that are already inclined on putting their name on everything to add their logo/signature to AI art and say "I made this" like writing under 100 words for a prompt is hard. So at the very least you can lean on the laziness of people to not go grab a watermarker program like uMark or just use gimp.

On a last note, stable diffusion already adds an unseen watermark that can be recovered and says the image was made by stable diffusion.

Edit: I don't have a horse in this race either. If auto decides to add a watermarking system and it is not forced, I won't go out of my way to complain about it.

[EliEron]

To be frank I have to side with the "watermarks are pointless" side. Anybody that cared for the honor system would already be transparent about their art being AI generated, and anybody that is trying to pass AI art off as traditional art is obviously not going to add a watermark.

At the end of the day watermarks like this are pointless unless they are mandatory, but due to the nature of open source software and open models, making them mandatory is impossible, thus making the watermarking option completely pointless.

Also if artist are mad about AI artists "stealing" their art currently, do you honestly think they will be less mad if people start plopping their own watermarks onto the art that they think of as stolen? Watermarks are usually used to assert ownership and mark copyright, adding that to an AI image is just going fan the flames even higher.

[RoyalShooter]

NovelAI watermark is just an example, my main idea is to put a visible watermark so people can tell this is AI art at first glance.

You don't like AI art, there, turn it off and you wont see it.

My goal originally with this idea is to help with the current controversies and let more people to start accepting AI art before every community starts to ban it.

Also if artist are mad about AI artists "stealing" their art currently, do you honestly think they will be less mad if people start plopping their own watermarks onto the art that they think of as stolen?

The idea of this watermark is not for the artist but the audience, what I found from one of my recent posts is most of them care more if the art is tag AI or not. that's why I have an idea to make a watermark(overlay) on it.

but I do agree "watermarks are pointless". This is unlikely to do a huge impact and make everyone happy. But I think it may still help with the current situation. Seeing you guys react it seems like making this into the repo really didn't make sense and no one will use it. Then my next step should be thinking to start with something like a Tampermonkey script to do the job, which with a little twist will also be able to support other services like NAI.

[EliEron]

My goal originally with this idea is to help with the current controversies and let more people to start accepting AI art before every community starts to ban it.

Don't get me wrong, I don't fault your intentions, they seem noble enough. I just honestly don't think it will achieve much. If you really want to make a PR I certainly won't stand in your way. But I'm not sure it will be worth the effort, especially since there are people far more passionately anti-watermark than me that would likely complain about it.

Then my next step should be thinking to start with something like a Tampermonkey script to do the job, which with a little twist will also be able to support other services like NAI.

I think a script that tries to detect AI images based on the metadata and automatically watermark them would honestly probably work a bit better for your goal, though it would not be perfect. As any re-encode of the image would in most cases discard the metadata you would be looking for. And a lot of image hosters re-encode images, especially if you upload them as PNG, since that is a lossless format that takes a lot of space compared to JPEG or WebP.

Ultimately there is no real way to truly prove whether an image is AI generated or not. If somebody really wants to fake it they can do so with relatively little effort.

[Heathen]

Ultimately there is no real way to truly prove whether an image is AI generated or not. If somebody really wants to fake it they can do so with relatively little effort.

Consider this, AIs that evaluate if an image was made by an AI or not are easier, faster, cheaper to train and use. So I highly doubt it will get to a point where you can't tell something was made by an AI, specially at higher resolutions.

[cmp-nct]

To do that properly that's a lot of work for something you can already do way better.
Just use a script and imagemagick "convert" to add a watermark or whatever other effect you want to your images.
I'm sure there are also bulk tools with GUI available for that purpose.

Makes no sense to integrate such a feature into this repo, it's barely needed, takes a lot of work to do properly (antialiasing, alpha channels, various input formats, location, storing, multiple settings, etc).
However, with a custom script nothing stops you doing it, I just don't think it's a smart way to solve that task

[RoyalShooter]

I thought to use a custom script at first too, but isn't it only work while generating the art, can I do something like putting a button into the script to change the art after generating it?

[cmp-nct]

Though let's be realistic: people can not ban AI art.
Even now it will be very difficult to spot the difference and the clues/mistakes made by diffusion models are not going to be there for much longer.
Within a year I predict a model to be released that was not trained on cropped images and will be able to correctly draw hands, fingers and toes.

Anyone who wants to ban AI, that's a lost war. Those who ban it now have policies in place they can't enforce.
Even today, with enough patience you can produce flawless images no one can identify as AI art.

[Heathen]

Anyone who wants to ban AI, that's a lost war. Those who ban it now have policies in place they can't enforce. Even today, with enough patience you can produce flawless images no one can identify as AI art.

Make me a sign that says, "This was made by an AI" with a person holding it. In 1080p at least.

You cannot use anything other than AI to do it.

Besides, Midjourney, Novel, Dall-E, all of these have this very peculiar style that if you see them enough, you can tell which AI made what picture. Don't underestimate the human brain's ability to detect patterns.

But the point in image boards is not to ban AI image, just to let users toggle viewing it on or off. Don't want to see AI art, turn it off. There you go. Problem solved.

[moongear]

This was made by an AI

Two years later: Hands and feet are still a major challenge.

[thesweetiger]

I was looking also to put a kind of watermark, Just the watermark wil mention generated by AI , sometime I send some of them the person know and he resend without mentionning , so I want to put a guard on it

[abzaloff]

You can add a watermark in almost any graphic editor. And in batch processing, too. There is no need to make suggestions to overload the already overloaded interface. If you need a watermark, don't be lazy to run images through the editor. The developers, I think, have enough work to do.

[Woisek]

Sadly, it looks like everybody still has not understood the purpose of an image generator.
Those things output a work file and not a finished product. Everybody serious knows, that after an image is generated, it has to be edited and finished. There is nothing like a perfect finished image out of a generator. Never. Not for the really professionals.

So the question about adding a watermark yes or no isn't even raised at this point. It simply has not to be placed on this stage of production. And even less automated. Depending on the motif of the image, it may be possible to place it elsewhere, adjust hue, saturation and/or lightness. The time for placing a watermark is at the very, very end of a production (if it's really needed) and then the image generator is already long gone by that time.

In conclusion, no, it absolutely makes no sense to add such an option, as it's not the place for that at this point in production. Every one who don't understand this, should learn how production and finalization is done first. And if they still don't get it, they deserve it to place a watermark by hand on an unfinished, sub-par image.

[moongear]

Sadly, it looks like everybody still has not understood the purpose of an image generator. Those things output a work file and not a finished product. Everybody serious knows, that after an image is generated, it has to be edited and finished. There is nothing like a perfect finished image out of a generator. Never. Not for the really professionals.

So the question about adding a watermark yes or no isn't even raised at this point. It simply has not to be placed on this stage of production. And even less automated. Depending on the motif of the image, it may be possible to place it elsewhere, adjust hue, saturation and/or lightness. The time for placing a watermark is at the very, very end of a production (if it's really needed) and then the image generator is already long gone by that time.

In conclusion, no, it absolutely makes no sense to add such an option, as it's not the place for that at this point in production. Every one who don't understand this, should learn how production and finalization is done first. And if they still don't get it, they deserve it to place a watermark by hand on an unfinished, sub-par image.

You are not completely accurate in your statement. I mean, if you are implying that by not manually refining a generated image with an external tool leaves nothing other than something that is 'sub-par', that is just not true. Sure, there are times I drag something over to Phtoshop for a some refining. But more often I see a finsihed product after having spent hours modifying the prompts, tweaking other settings, trying different models then finally look at it and say "WOW, that's incredible! It's finshed."

[moongear]

I found this because I, too, am looking for a way to add a signature to generated images. I disagreee with pretty much all (actually, probably all) of the negative comments about having this ability built into the GUI. So, I'll break down a few:

@Heathen wrote in part: The other problem I see with adding a built in watermark is that it gives the people that are already inclined on putting their name on everything to add their logo/signature to AI art and say "I made this" like writing under 100 words for a prompt is hard. So at the very least you can lean on the laziness of people to not go grab a watermarker program like uMark or just use gimp.

Haiku has fewer than 100 words, and they're usually signed. But that aside, some of my generations - with let's say around 100 words - took hours to refine into the finished version. If I choose to put my signature on something, it is because I have earned it. So many comments on here refer to "Art" this and "Art" that. This is an artform, for sure. And artists throughout history have signed their works so I fully support an automated solution for achieving this.

There are pitfalls associated with using soemthing like Phtoshop. More on that later. Plus, I have generated more than 17,000 images in tha last month or so. And that doesn't include the hundreds, probably thousands, of times I have clicked the Interupt button. 'Laziness' has nothing to do with it. It boils down to what's practical and what is not. Again, more on that later.

@ZCaliber wrote in part: I think the best analogy I can come up with is that we are all creating recipes for a cook. We are creating a method for creating an image that, hypothetically, can be recreated exactly by someone who can duplicate the 'recipe'.

Maybe I am confused at what this statement is intended to say, but I am not creating so others know how to 'cook' up a particular image style. I am doing it for me first and for the public second. Besides, the 'recipe' would still be in the prompt regardless of whether or not a signature (watermark) was implanted onto the finished product. It seems this is likely true for many Generative AI Artists. We are not doing it so YOU can do it, too. Although, the nature of many sharing websites helps create that beneficial byproduct by including the embedded prompt and other information into the posts.

Don't get me wrong, I have no issues with someone taking my 'recipe' and making it their own. Just as long as they don't generate the exact image then call it their own. However few words the prompt may have, that's despicable.

---

There are others, but those two I found personally to be the most thought provoking.

Generative AI Art is a skill, a talent, a love and a passion for many people within the culture. I completely agree with @quasiblob that this feature would be a nice addition to the workflow. I like the idea of a transparent PNG sitting on the sidelines that can be overlayed with the click of a button.

Unlike Photoshop and other resources, this could retain the metadata in the original generation. Which, by the way, is something that so many people might not understand about Photoshop. I use Photoshop extensively and no matter what I have tried, the metadata, aka 'recipe', always gets removed. I have to use another application, TweakPNG, to put it back after using Photoshop.

In summary, I completely disagree with the nay-sayers and the 'logic' behind their opinions. I can see a legitimate use for this kind of tool, just as there is a legitimate use for implanting and other tools. My vote is absolutely, yes, or perhaps as a third-party extension. Soon, I may even begin applying copyright laws to select generated images with my prompts and their accompanying seeds. Yeah, stick that in your nay-sayer bag of discussions. It's happening.

I will leave you with a quote from Attorney Adam Adler:

… proponents of AI art don’t have to look very hard to find the required creative contribution. The most prominent AI works are generated through trial and error using specially crafted word prompts. For example, Jason Allen, the winner of the Colorado State Fair, spent 80 hours crafting the prompts he used to generate the art and tested over 900 different prompts before settling on the winner. Given the sensitivity of AI art generators, one could argue that the selection and refinement of prompts (at least as they are used today) involves significant creative work, analogous to placing a camera or framing a shot. And because a human’s prompt selection informs the creation of the entire work, there would not be any obvious way to disentangle the creative and non-creative elements of the work

You can read his entire article on the subject here:

https://www.escapistmagazine.com/the-horror-of-ai-art-in-copyright-law-other-legal-horror-stories/

And there's this:

https://www.smithsonianmag.com/smart-news/artificial-intelligence-art-wins-colorado-state-fair-180980703/