You don’t make the creative decisions in the place of textured brushes, they paint where they will. The result is also often subpar, requiring a lot of work fixing it.
Not unlike what a fast line art tool would be.
Then we go after them. For that happen we need legislation. There’ll only be laws if we make compelling arguments for them, and right now we’re losing that fight. Gen tech is shiny, just saying “no it’s terrible” won’t cut it. We need to dismantle the notion that labour-theft and copyright breaking are required for them to exist.
Because they will exist, whether we like them or not. Wishing them away won’t make them go away. But we can regulate how they’re developed and employed.
The line art tool does not “all the job”.
The job is primarily made by the artist, who’ve drawn the sketch.
It’s not a prompt. No good quality sketch, no line art
The reference to 3D tool is because you’ve wrote that this kind of tool should be implemented instead of line art tool… If you didn’t refer to it, I didn’t refer to it
And yes, a 3D model then you need to draws over it… like the sketch: I need to draw it.
No difference.
Except that for 3D stuff I just avoided all the hard work to compose and draw my character with the right proportion according to view angle and character pose
Traditional artists, that work with paper and pencil, may not like this kind of artists, that just does over draw
From the moment you use a computer, you automatize things
But if you take a look on Digital Art, when you draw a bezier curve with assitants, this is a kind of automation.
The brush engine itself automatize a lot of computations to generate the pattern drawn on the canvas.
Using a “noise” filter? result is an automated stuff.
Using a “screentone” filter layer? result is an automated stuff.
Using a “transform layer”? result is an automated stuff.
I’ve never considered myself as an artist…
To my knowledge, Krita’s fondation is still against gAI for which:
datasets are built on stolen artwork
results are 100% generated, from nothing else than a prompt
To my knowledge, Intel already in the past acted as a sponsor for Krita to let brush engine performances be improved with use of SSE instructions I think…
The result of the line art AI would just be more tedious than correcting the brush strokes.
That’s not true. I’m sorry, this claim is based on your opinions. What’s actually going on is the opposite, that we are winning. Either way, that doesn’t have to do with this specific project.
The difference is that here you dont draw the line art anymore. There is a difference.
You still have to decide on how to pose it. The program doesn’t do it for you. You still have to draw. The 3d tool with this have nothing in common.
The transform layer doesn’t automate?? It literally just scales your art.
There are halftone & screentone prints. They have been made previously by others, now it’s used as a filter. Nothing in this is automated?? Nor does it have anything in common with the Line Art project.
Where did you get this information from?
That’s good, but that doesn’t mean they fully support artists. Again, you ignored the fact that Intel has sponsored SD.
You don’t make the creative decisions in the place of textured brushes, they paint where they will. The result is also often subpar, requiring a lot of work fixing it.
Not unlike what a fast line art tool would be.
Your art is amazing so you probably know better than anyone else on this forum that the degree to which you don’t make decisions with a brush is so much lower than what you’d have with this fast line art tool, you get to control the brush behavior yourself, you can make your own textured brush from scratch to get the desired outcome, it’s a simple tool that you still have to operate with mindfulness. Fast lineart will surely be built with the goal of improving it until it will be flawless, and while it’s true that the devs are asking for consent for their training data and not exploiting works that are already available online, this is happening with Intel funding during a time in which there is no legislation whatsoever, so what’s stopping them from going all in at this point? Yes, Krita devs have earned out trust so far but this feels like it could end with a bad outcome, especially after they stated that they were against genAI (which this is technically not but let’s be real, it’s going in that direction).
I’d love to hear more about it, no sarcasm intended. Can you point to any new laws regulating datasets?
You can’t cite all the wide happenings in the field to support your arguments then turn around and say it’s about this project in isolation.
You also didn’t engage with any other points I made. Earlier in this thread you complained your points were being ignored, so I tried to engage with each one you raised. But this is becoming a “you said I said” of handpicked points that leads nowhere. If the next replies are in the same line I’ll just bow out because what’s the point?
@nedow
I understand your concerns and share them. But if it has to be done (an art tool using ML, and as I said before I believe it has to be done to bring change), wouldn’t you rather it be by a small independent team which can be pressured by their userbase into doing the right choices?
Cold and calculating, I know, but I’m not one to put my trust blindly into anything. Past is context, it earns good will and lets you predict with some confidence how they’ll act, but is no an assurance someone or some entity will keep behaving ethically in future. What’s stopping them? We are. Because the power imbalance isn’t so great they can afford to simply bulldoze over us.
I’m far more concerned by the open source code being appropriated by the genai griftsphere and used in some awful new contraption. In this aspect I think it’s a substantial risk, but I also suspect similar tools would be developed no matter what, and again I’d rather it be by a group that can use it to emphasize the artists point of view.
Correct me if I’m wrong but from what I understand, generative AI is in essence algorithmic photobashing. It trains itself to algorithmically recreate each training image as accurately as possible, and thus it can then generate all of the trained images plus all possible interpolations between them. Additionally it is also trained to associate the images AND patterns within the image with words. Thus it can also pick individual patterns from within their algorithmic soup of original images and interpolations thereof and arrange them to make new images in response to prompts.
This implementation seems fundamentally different than that in that it’s just being trained to average sketch lines into singular lines and discard extraneous info, so fears of this being another Mr. Steal Yo Art like generative AI are completely unfounded, thankfully. Even if this was unethically trained off stolen images it wouldn’t be capable of creating images on its own.
HOWEVER- I do have some concerns that this could be used to help disguise AI proompting scammers. Up until now they could only pretend to have made sketches and lineart by running their junk through basic image filters. This could potentially be used to make those fakes seem more genuine. I have concerns about other less savory people using this and the gathered data to improve their generative AI models too, since this is going to be open source after all.
Also like galaxcolour so sagaciously pointed out, this is less of a tool than a thing what does all the work for you and then you clean up after it. In that respect it’s too uncomfortably close to Gen. AI. Unlike generating flats, which is a tedious busywork job that needs zero creative input, lineart is a fundamentally expressive task that requires you to think, feel, and train yourself to use your tool as an extension of yourself. It’s also, unlike the example of flats, a major component of the finished image, especially if there is no color intended to go in the work. Really it’s far too important to foist upon an arbitrary black-box neural network. I say this as someone who is far too lazy and attentionally challenged to do anything but spit out seat of the pants scribbles in one sitting. There’s also the fact that your sketch has to be clean and well defined for this to work, by which point you’re already most of the way there anyways. And as much as people might find it annoying to ink their sketches I can’t help but feel that having to clean up after the AI would be even moreso.
I am not enthused by this project. Handmade linework is beautiful and it communicates part of the artist’s soul and body. It is a part of their fingerprint. I don’t think that’s really a good thing for a dumb brute-force algorithm to be handling for us.
reverse engeneering “proof” of non AI AI generated imagese is already out there:
I really doubt this tool is going to help people steal art or fake “real” art, it’s just going to help people who don’t have the time or physical ability to do proper digital line art over their sketches.
Thank you, @Grum999, @Celes, @BeARToys and @jointri, for your contributions, contributions that I currently lack the strength to make, you speak from my soul and also seem to have read this topic in full, and also seem to understand the background and possibilities, as I think I read from your posts.
All I know is that fear has never been a good advisor, and that true strength lies in knowledge and the ability to evaluate and weigh things up. But in no way does strength lie in denial or prohibition, these have always been the mechanisms of those who do not join the ranks of the deniers and disregard every prohibition to their own advantage, as they can gain a head start in times of prohibition and thus an immense advantage over those who prefer to see everything prohibited.
It is much more advantageous for everyone, and therefore also more important, to accompany things from the beginning and to work out good regulations in the interests of everyone, regulations that allow fear to be taken away, so I think the approach taken so far in this topic is good.
And if anyone wants to know, no, I won’t be using this technique if I’m ever able to paint properly again, I’m a painter and haven’t needed any line art so far.
Am I less of an artist because I use the colorize mask tool? Not even AI just clever algorithm but of course automizes flat colors. People could take line art or clean sketches from the Internet and “auto color” them. Artists could get bad at flat coloring because they don’t do it anymore. What stops developers to improve it so it even shades.
Again. Not AI. But the same arguments against the fast line art tool can be made (and actually were made) against the colorize mask.
I think it’s really important to distinguish how much agency is taken away from the artists and how much of the completed work was done automagically. This can vary on a piece by piece base, with tools like this. I personally don’t consider line art a finished artwork. For me it’s only a fraction of the work.
More like questioning the utility of this feature and for who is it for, it’s a fair criticism to make.
People defending this can’t see the bigger picture at all.
More criticisms that hurt less fee-fees:
What’s stopping people from inputting artwork that isn’t theirs into the model? Krita has explained nothing on how they are going to curate and audit their inputs and guarantee a clean dataset.
If a bad actor inputs artwork that isn’t theirs without consent of the original artist, what then? Will Krita be held liable or not? And if so, what will Krita do? Rollback the feature or stay unethical? Didn’t they make a public statement that AI has no place in Krita and that they are on the side of the artists? Isn’t this contradictory?
They are also not explaining that they will and how release the dataset for community audit. So much obfuscation and a lot of “trust me bro”
There’re some few “new comer” who’ve joined the forum some minutes/hours ago here just to tell how this functionality is bad and (for what I interpret because of my English but it’s maybe only me…) how we shouldn’t trust the dev team…
No background, no artwork, nothing more to present than a “I disagree”.
Not sure if it’s the same people behind different accounts or not, but users who never take part about forum suddenly decide today to create an account here and tell things the hard way
Hopefully I’m not an artist
So I don’t feel concerned by what could be a real artwork or a true workflow, I draw my stuff as it comes
This is wrong. If this was even remotely true, AI would be the ultimate compression tool as the training datasets are many terabytes large, and yet the final AI program is a dozen GB at most. AI cannot recreate the training data or ‘photobash’ unless it’s improperly trained.
To keep things on topic, if you dislike this feature ‘because it’s AI’ then you probably don’t know enough about either this feature or AI to have such a strong opinion.
If you dislike this feature because it’s ‘lazy’ or ‘taking a vital part of the process and automating it’, well I guess I’m right there with you and agree. I wouldn’t be so against the Krita alternative to Adobe livetrace if it wasn’t taking dev time away from better things.
Personally, I may be that rare artist who actually likes or at least doesn’t mind doing the line art myself. That being said, this will have zero impact on krita performance if you don’t use it, so it’s purely optional and thus a tool. Let me give you a situtation: you’re a system administrator and write up a quick and dirty perl or bash script. That didn’t kill jobs, that made jobs for those who do the human work much easier. AI could be used for good like how it can be used for evil. I just don’t see AI as some extremely tribal absolutely good vs absolutely evil.
Let’s just ignore the personal attack because this kind of thing sucks the oxygen out of the room. I’m fine, and personally okay with anyone thinking I’m an idiot and a terrible artist. Responding would further derail the thread.
I’ll just point out that the chainmail example was just an example: A stand-in for “painstakingly time consuming and detailed work”.
The disagreements aren’t without merit, I’d very much like to hear how the dataset would be audited. It would be a positive if it were disclosed, it sets a precedent.
I’m also concerned about misuses of the tool.
I’m just not ready to decide the tool is being developed in bad faith. I think the devs have the right to address the concerns as they come without being considered already guilty and face possible harassment for it.
the main post has made it onto reddit, so expect a lot more of this, especially people coming in to make knee jerk reactions and not actually READING the main post and all the explicit explanations the OP makes about how this process is to work and is to be trained
Call me cynical but I don’t understand how people can be so thrusting. They’re being backed by Intel. They were explicitly anti-AI a couple of years ago, except thanks to MONEY now they’re not. What makes you think that in 2 more years they aren’t going to scam their userbase and claim that all art made in Krita is going to be used to feed a gen-AI algorithm, like almost every other big company is doing today?
