AI turns photos into cartoonlike paintings

Originally published at:


Art imitates life. . .better than ever before!

1 Like

Seeing the smoothing and flattening effect in action with video its effectiveness is lessened, doesn’t look nearly as good as with a still image. I imagine that with some extra time tweaking settings a better effect can be achieved

1 Like

Impressive. Not quite up to Bob Sabiston fare - yet.

This is impressive and I’m sure it will improve with time. But even I, with basically zero graphical art ability, notice the difference between simply reducing a reference photo to a cartoon effect of surfaces, shadows and angles, and the artistic license taken by skilled animators. I’m not sure how great animators and other graphical artists make those stylistic choices.

I’m not saying a machine intelligence is fundamentally incapable of reproducing Ghibli-esque art, but this ain’t it.

If you showed this to me without telling me it was algorithmically generated, I wouldn’t know it wasn’t directly drawn by a human artist. But I don’t think it would be any more inspiring as art. After all, if a machine intelligence could create that kind of art, then it would still be art, albeit with the intelligence’s architect as the actual artist and the machine intelligence as the tool use to crate the artwork.

Even this is art, but the state of the tool isn’t up to Ghibli-esque quality IMHO.

Still pretty neat.


If you use VLC Media Player, you can do something a little like this to watch any video in real time as a cartoon. Go to Tools, then Effects And Filters, then Video Effects, then Colors. In the Color tab, put a checkmark in boxes for Gradient, Color, and Cartoon. Voila.


The main (first) image on the BB article is misleading. That is a frame of the anime Garden of Words (言の葉の庭) by Makoto Shinkai; it is not an example of the AI output/result. This frame is visible at around 1½ minutes in the film.

It seems that that image was taken from the last page of the supplementary PDF, where a comparison is shown between human-drawn cartoon/art images and AI output/results for a similar scene.


It’s interesting, some of the pictures - and most of the video - don’t read as “cartoon” at all but “photograph with filter.” Because animation/drawing rely on not just simplifying, but selectively simplifying, which is very difficult to imitate without a brain directing what elements get extra detail*. (And that’s leaving aside style whereby forms differ from simplified photorealism.) That the system seems to throw in some artifacts doesn’t help.

*Though in theory an “AI” system could do it with enough examples - by copying the work of humans where the kinds of simplifications are appropriate to the specific subject. But then that becomes really complex - the system has to “recognize” most objects in the image for it to work.

Ha, that was the only image that I thought was really successful and impressive. Well, it explains why the rest of the images don’t live up to that…


Looks like Waking Life rotoscoping.

1 Like

I wonder what an AI-produced cartoon image would look like when run through again.

Machine learning is not AI. This is just a photo filter using a generated algorithm.


Yeah, this is basically Photoshop’s “Posterize” filter in realtime.

Android “Prisma” app can do pretty much that, plus a lot more.


This topic was automatically closed after 5 days. New replies are no longer allowed.