Greetings Developers of Krita and Krita’s community as a whole.
Can we discuss a moment about a topic ?
I was wondering. Is Krita using AI technology at all as a component of its architecture or one of its development tools ? Do you think creating/using some AI frameworks could help improve Krita anywhere at all ? What do you think of deep learning and its usefulness in Krita’s development ?
Greetings Developers of Krita and Krita’s community as a whole.
For you specifically, how about getting a good PC instead at some point.
In my impression, “machine learning” is the only thing that is not in the plan.
The performance really has to do with other things, for example on Windows Krita works fine for me, on Linux it works very well, quite fast.
I’m just curious, really. I know Krita is pretty fast in many parts already today.
I’d like to hear thoughts about AI/machine learning/deep learning’s usefulness in Krita’s development and what the developers think about it. Basically, I want to know if it has been explored before as an option and if there are benefits to use this tech in Krita’s development.
Ah… Thank you for the info.
Now the discussion can be dismissed.
I’d like to see an AI system used to answer problems about graphics tablets but it would probably melt or resign or self destruct.
It is for now. Who knows what future holds.
For now AI seems to be mostly aboit pattern recognition. So selecting skies or objects in Photoshop or removing noise/ sharpen in Topaz. For the most part AI is not that important in workflowand more of a commercial buzzword. Masking skies in photos is useful though.
Adding a -1, as am against AI to the point wouldn’t use krita if that was added, but, as TwoTwo has quoted, it’s very heartening to see it on the list of ‘stuff we shouldnt do’.
AI literally could make genuine artists obsolete, or, at the least, genuine artworks less seen or shared. The most recent conversations I’ve seen by professional artists questioning AI included comment that possibly only personal offline art would end up being the least affected by AI e.g. local commissions for family/friends etc.
You defined a precise context in which it’s not useful at all.
More relevancy with a context in which it would help is better.
Like for instance, maybe AI could improve the speed of filter rendering in Krita, or stroke speed.
In such case, it would be useful vs useless when it does everything by itself, which is totally pointless, because we’re not looking for that here.
That’s why precision matters, and you gave your context, so it’s fine. I would also add a -1 in your situation.
But if AI is proven to increase significantly a lot performance while being not hard to implement thanks maybe to a particular framework, then it will be a plus 1 instead.
That’s why @raghukamath said that, for now, AI is not so useful, but who knows in the future ?
@novames00 I’d put general feelings about AI and the concerns many artists have regarding it (which also includes boycotting it, due to those concerns), but, if talking about krita functions more specifically, the projects document containing the quote about not including machine learning also covers about continuing bug fixing, which I understand is one of the most important things for keeping krita running well and as part of any code improvements or additions. There are many time-tested remedies that can fine tune, clean up code, etc … jumping past those things and straight to AI, which is directly linked with being anti-art/artists, and obviously would impact on the future, I don’t feel is useful or morally considered.
There are no doubt other creative softwares incorporating AI on windows, but, particularly for linux, there is only one main painting software, Krita, and it’s probably more likely that linux users wouldn’t want AI involved within Krita, with privacy etc being important.
You probably think about programs that can create entire artworks automatically. That’s a very specific use case that is unlikely to ever appear in a drawing program because why would it? Although it could have some interesting uses for the art equivalent of fast prototyping.
More generally AI is just a tool in an artist tool box just as 3d rendering, 2d shaders (which are basically already available in krita) or procedural generation is. AI definitely can improve existing tools like transform or selection tools.
I see a lot of people often overestimating what AI can actually do though. Sure, there are some pretty impressive examples but these are mostly a tiny minority. Most AI isn’t even implemented in every day software because that’s unpractical. The AI was used to create heuristics and those are used instead.
And AI also doesn’t automatically mean that privacy is violated in any way. Simple AIs can run just fine on the local machine. Also, you don’t need a fancy AI to spy on people, any malicious krita plug-in could do this much simpler.
Anyway, I still think Krita should focus on other things right now, which they do. Just wanted to say that AI doesn’t have to be anti artist per se.
I’ve lifted only a few points into the quote as so much could be said about these things. I wasn’t thinking about artworks being auto-created, but about AI itself. AI isn’t harmless, as, although it can aid in various computer processes, it is a large part of IoT/data mining/any track and trace, etc etc … so many ways AI is used in ways that increasingly violate privacy, and intend to (e.g. digital money system I keep seeing referenced, and, with e.g. the cancelling that happens online, it can be seen how problems could arise).
Some other examples … online posts that are bots/AI, and can push certain points; AI doing online customer support (or blocking avenues of enquiry) can be an issue; robots brought into the work force increasingly; algorithms that block artists’ work (or videos) from being easily seen (I’ve seen increasingly comment, and misery, about that), and also far more data is taken nowadays, and sold/used, due to AI.
I have a fundamental issue with things increasingly being automated through non-human routes, and I’m not sure auto-trusting government programs is always wise. I reference these things that seem to have nothing to do with applying AI creatively as my point was very much a moral one, and, whether artists or not, many are concerned about what AI is and could increasingly do.
As you say, Krita already contains 2d shaders … an example of everything functioning, without AI, plus, as the Krita document references, there are some ‘nasty build-time dependencies’ regarding AI, so it sounds like it could bloat Krita and possibly go against the GPL. The document also says that it’s something Intel’s been pushing for over the last 2 years, plus they are considering becoming a corporate sponsor … jpeg-xl being included has an attached note that says intel really wanted it, but I’m really impressed, and relieved, that Krita haven’t brought in the ML Intel seem to really want.
Most things you listed aren’t even AI in the sense a computer scientist would call it more on the level of a AI of an NPC in a computer game, like bots doing help desk answers, this is one of the primitives AI i could think of mostly because the missing intelligence part.
Anyway I understand now what your concern is. Image recognition for surveillance snd stuff is a pretty shit use and a dangerous one, and although it’s just a fraction of what it’s used for I understandust a when someone wants to stay away from this as much as possible.
Well this conversation has taken direction quite far from helping optimize algorithms or finding new ways for algos to big brother watching you.
Since Krita has open source I highly doubt adding some data driven/machine learing alogrithm would be a problem even for linux community. AI doesn’t mean surveillance per se, you also don’t need AI for surveillance so these general doubts you have don’t seem to stop anyone from using Krita or working on it even now.
Now a little bit back on track with the topic.
One part where AI might eventually help Krita is the code base, the AI helping with scripting/debugging is now a pretty big topic in development and it’s already at a useful state, I don’t know how much Krita devs are into it but at some point I think it’s safe to say that someone will probably use it, it’s on a way to become industry standard, it’s already a test feature now, soon it will be an optional tool and after that it the dev environments use it to replace current scope, like it or not.
One area where it could maybe help Krita if someone felt adventurous, Krita doesn’t have and doesn’t plan to have one of those simulation systems for painting (for example to simulate watercolors, …), the system would be probably a pain to develop but maybe with help of machine learning there could be some nice middle ground.
This is something I tried already at work and it really helps finding shit stains in code before they become an issue. I didn’t try automated input testing by AI yet because that only seems to work with certain UI framweorks but that’s definitely an interesting and useful application.
Thank you, and I agree with you also; such things can do a lot of harm, and are entirely unnecessary. People shouldn’t be treated in such a way.