Line Extraction and Painting

We have the technology to extract line art from rough sketches.

hi.cs.waseda.ac.jp/~esimo/en/research/sketch/

We can now "paint" image sketches in a style of our choosing.

github.com/lllyasviel/style2paints

We are only a few steps away from customized waifu creation and modification

Try the demo out for yourself:

paintstransfer.com/

Other urls found in this thread:

kvfrans.com/coloring-and-shading-line-art-automatically-through-conditional-gans/
github.com/alexjc/neural-doodle
youtube.com/watch?v=_RM1zUrY1AQ
youtube.com/watch?v=F6ex4pogilA
youtube.com/watch?v=fu2fzx4w3mI
imgur.com/a/hW4LS
hi.cs.waseda.ac.jp/~iizuka/projects/completion/en/
hi.cs.waseda.ac.jp/~iizuka/projects/completion/extra.html
twitter.com/SFWRedditVideos

Here is another similar effort that allows hand-painting in the case that the reference image does not produce the results you want.

kvfrans.com/coloring-and-shading-line-art-automatically-through-conditional-gans/

The results are generally high quality and can be used for things like manga colorization in real time.

Here is a complex scene that is stylized using the original artists' color pallet.

And this is the prior line art.

This is seriously amazing. Hampers my spirit to learn drawing somewhat though

You can see that a simple cartoon without inking can be transformed into a full line art drawing that is ready to be painted with style transfer techniques.

Yes. This is the state of the art in neural network based colorization of 2D artwork.

If we combine this with approaches similar to those outlined in neural doodle, we can achieve amazing results.

github.com/alexjc/neural-doodle

How did you get into this? Where can I read up more on this, outside of the links you already posted?

As an artist this makes me upset.

What exactly makes you upset? When I see I just get scared about creative work becoming even more devalued, since it will be so simple to create seemingly high quality of work

As someone who would like more art of particular styles, this makes me happy. Still not where I would like it to be yet, but a good direction.

Keyframe or In-betweener assistance would be fun.

"Scalable" (Detail-sense) vector graphics would probably help too. (3D-Hairmodel -> variously detailed 2D-Projections in a specific style).

As far as In-betweening is concerned, you might have already seen these: youtube.com/watch?v=_RM1zUrY1AQ though I have to admit that the colorization is a lot more impressive. Though my experiments with it haven't been to stellar either

I strongly feel that this is an insult to life itself.

drawfags on suicide watch

Useless trash. Maybe useful for chink chong animators and devaintart tier shit artists.

These are new solutions for solving the artistic style transfer problem.

If you want to learn more, you can look up artistic style transfer and check out the latest SIGRAPH conference videos.

For instance, you may find this video interesting.

youtube.com/watch?v=F6ex4pogilA

You still have to be able to draw the sketch first.

That's kinda depressing, actually.

I wouldn't say that's ready to be colored by a long stretch. If you spend more than a few seconds looking at it, you'd see that there is a large amount of touch up required, to the point where you might as well have traced it manually.

No

If you look at the interesting work being done in stylistically interpreting rough sketches, you can see that the artists job is less cumbersome now than ever before.

youtube.com/watch?v=fu2fzx4w3mI

Although this is a simple video using a somewhat unattractive style, there is potential for improvement and style variation.

Drawing/painting = solving the problems by making decisions
Neural shit can't do this.

Best thing I could make so far.

It works best if you have a picture of the same pose as the sketch you are uploading

It's still a very, very long way from creating any interesting line art or paintings
we had this discussion on /ic/ before

Neat how it can interpret the color reference image and make a coherent picture. Color me impressed!

...

>coherent
bruh

Skin color and eyes are really hard to do.

Ruined. But at least it works with simple images, so that's a positive.

This one is pretty good.

I'm going to need both upper left corner image and right image, right NOW to test and work with the line art myself to see if I can get similar results. I'm not artistic but there's a lot of images that might turn out nicely after some tinkering.

This is clean lineart. You can achieve better results even with mouse.

Tinkering is the best man imgur.com/a/hW4LS

pretty fucking dope.

...

We can remake her. We hve the technology.
Not that I would.

That moment when you regret not having saved more sketches.

Someone do a manga still!

Come back to me when I can google a random image of a girl and have a stylized sketch of her done in an instant. What's the point of something that draws for you if you need to already know how to draw in order to use it?

This feels very wrong to me. What's the point of artists now? Only drawfags will be relevant while colorists and full package artists will turn obsolete.
Once again technology doing its best to remove human soul from all artistic work.

It really doesn't look that good.

Is this bait?

I added some sketches and line work to the album imgur.com/a/hW4LS

I like this one

Thanks for the tip user.

>ones on the right are absolute shit, completely unusable, might as well trace it instead
is tho bait?

That's the joke.

I keep trying but I get a lot of image bleed out around hair sections. Am I doing something wrong or do I just need to be more careful not to go over the line borders from hair and back ground?

>PS can do the same things but better for decades

That one really looks on point.

I stopped trying to add more colors outside of skin color, it doesn't lead to decent results most of the time. A pity it doesn't work out for the head in that image, otherwise it's a great fit

Yeah, I was messing with mostly NSFW panel images of my favorite series and all of them were coming out a mess, so I tried something simpler and still couldn't do it. I could see this being very effective if I had something that was more accurate than a mouse to ensure I didn't go over borders and edges.

unfortunatly the original already had a low resolution.

As a scanlator this gives me a boner.

Gambatte AI-chan

good thing im only focusing on learning to draw line art,

Until someone comes up with a function that takes an image as input and gives how beautiful it is as output, the best ANNs like this can do is emulate something that already exists and has many examples.

Sad how much better it is coloring my lineart than I am.

Are you using layers correctly? Start with light colors, then get a few tones darker are you reach the edges. That's what it basically does itself.

nice

This is really fucking neat

If you get bleeding in your image, you can use the color function in the top left hand corner to tell the program where to put the colors.

...

Feels like I'm cheating.

Talent be damned!

I did exactly that, I just didn't bother correcting it for long.

...

What's the source on your reference picture? I want to try this one out

all is doing is applying a filter and blurring the image. this thing won't do you any good if you want a colored pictured that is sharp and cares about the lineart

...

I used the the Mio picture (Upper left) for color reference and main b&w image of Takahara Ayumi

...

Damn

Mmmmm... Computer colored Christmassy

good stuff

...

...

If you use a reference image with a white background, you can keep the white background from the sketch image.

t-test

Yeah, this works well.

Are you working with Simo-Serra and Iizuka or just building on their work? I've been following their Colorization and Sketch Simplification work. Also anticipating the release of the code for Globally and Locally Consistent Image Completion (SIGGRAPH 2017) for some time, for possible applications in redrawing.

Whoops, out of order, this is the before picture and that's the after picture.

This looks like a content aware in Photoshop rip-off

hi.cs.waseda.ac.jp/~iizuka/projects/completion/en/

What is this supposed to show me? It does the same thing as Photoshop.

does anyone know why the finished picture is in much lower resolution then the source image ?

Impressive.

looking very kuro in there, kuroko
sorry I just had to post it

i didn't know, but it looks cute, like she (he) is in a park or something

...

It takes some effort if the sketches are a bit rough but I can't argue against the results.

You telling me Photoshop can fill in faces and objects like that?
hi.cs.waseda.ac.jp/~iizuka/projects/completion/extra.html

This took some serious tinkering, and the webapp keeps crashing.

What a time to be alive.

Sure is.

>Also anticipating the release of the code for Globally and Locally Consistent Image Completion (SIGGRAPH 2017) for some time, for possible applications in redrawing.

Can I decree jokingly that similar Threads should just use the title part "Rese/a/rch & Development".

We could do one on constructing the original analog Image from High-DPI-scans too.

Automatic or assisted soundword-removal and redrawing would be nice.

Pretty good.

...

Yes, this is just a rip-off of already existing technology that's used in Adobe products.

Can this be used as a command line tool?

...

why is she a cheeto