From that post, so context is easily discoverable here (I hate Discourse navigation, Discourse is the forum platform):

"It’s not a generative AI. It won’t invent anything. It won’t add details, any stylistic flourish besides basic line weight, cross-hatching or anything else. It won’t fix any mistakes. It will closely follow the provided sketch. I believe it won’t even be possible for a network of this size and architecture to “borrow” any parts of the images from the training dataset.

We will not be training the model on any of the existing datasets, or stolen pictures. All artworks will come from artists fully aware what it’s going to be used for. And our particular model will work better with special training data anyway, I believe. Maybe you’d want to help out with gathering the artworks - I will be making another post about that soon.

The calculations will be 100% local and offline. It won’t send the sketch image to any server to process and return the line art. I’m not planning to implement any networking functionality, and there are no servers planned either. It will only use your own computer CPU and GPU for calculations, the same way all of the other features of Krita do. It also won’t train on your images that you make in Krita. It won’t save it anywhere either, until you save it with Krita to your own device as usual, in a Krita file."

This being said, I don't think this is the correct approach. I don't think you need a convolution network or training. You would need some very carefully designed filters with parameters exposed to the user. Granted, this won't do as much of a "good job" at it, but the artist will touch it up anyways.

I think emerging technologies from now on won't be judged solely by what they make possible that wasn't possible before, but under what parameters/limitations they are morally acceptable to specific markets.

(And this isn't a strictly good/bad thing to me, it's a natural byproduct of a sufficiently advanced state of technology in the future being capable of automating every last productive human activity.)

I think the top post on the Krita thread does a pretty good job at setting their boundaries. Something that cannot replace artists: it will not "beautify" art, and stays close to the input, also it should not be trained on the work of unwilling artists.

Is the Krita filter trained on the work of thousands of artists?

It is trained on works of artists that fully knew it was being used for this.

I don't see the mental gymnastics here, the post you linked pretty clearly delineates the differences to AI in the sense it is used today.

As major differences I'd highlight: local and offline, so drawings not sent anywhere trained on artist work with explicit consent

In Krita's case, they claim the AI isn't generative so it doesn't add detail.

Whereas the AI today is trained on stolen work and often on the inputs as well.

Well, first of all, I don't buy the idea that there is a clear line between 'filter' and 'generative' AI. Even in the example Krita dev posted [0], you can see the 'filter' AI made up a bit detail (the way the girl's eyes look became different), it's just not as smart as so-called generative AI.

And about the privacy and copyright concern, what we currently have are:

Stable Diffusion: local and offline, but not copyright-clean.

Adobe Firefly: online, but copyright-clean (if we believe Adobe's claim).

So if we combine the better sides of both, it suddenly becomes okay?

[0]: https://krita-artists.org/uploads/default/original/3X/1/4/14...

Well, there is a clear line between a handmade filter and AI. But this is clearly AI since it relies on a filter optimized via automated training on images.

I think we are in agreement, I have used more descriptive wording to just clearly indicate what I consider as a filter.

(Edit: yeah, looking at the image I can see it clearly takes some artistic liberties. Even on the dragon.)