The previous tutorial was devoted to exploring various ways one could take a rough black line sketch made on a white background and automatically convert it into automatically filled solid colored regions. Which could also be vector colored regions if you wanted to do that.
Three different approaches to automatically do that were explained in detail in that tutorial. The Flat Region Colorizer ip op, the Vectorizer, or the Paint Synthesizer.
Of course one could also try to automatically generate the rough black line on white backing sketch itself using features available in the paint synthesizer. And again, there are many different ways you could approach trying to do this. So i thought i'd try out some different ones, and post the results here along with the presets i used to generate the final auto-colored regions generated from the fully automatic black line on white backing sketch.
Here's one quick example i threw together in a few minutes this afternoon below.
The V4 PASeq i used to automatically generate the rough sketch and it's auto-colorization is shown and attached below. It incorporates the Flat Color Regionize ip op technique as it's last action step to automatically fill in the individual regions defined by the rough sketch. You could of course use any of the 3 different approaches to ding this we described in detail in the previous tutorial post in this Group.
The sketch is generated based on 2 different basic concepts, working together to create the entire finished sketch. Based on the general ideas used when conceptualizing a rough sketch for a WPAP image. Generating global patterning while also incorporating local source image detail to work together to build up region structure with the complete sketch.
The AutoPaint action step shows one way to build up a stylized global patterning. It fracture the canvas using a set of 10 automatically drawn straight lines that radiate out from the center of the canvas at random angles.
And i used the same 1 pixel black hard edged paint preset i used to do my manual sketches in the previous tutorial. Except i custom programed the Path Start, Path Shape, and Path Angle control panels for that preset to automatically generate the global fracture stylistic patterning i was looking to achieve.
You can examine the control settings in those control panels in the paint synthesizer for that AutoPaint action step to see how i did that.
In addition to generating global region structure, i also wanted to add some sketch lines associated with the source images feature details. There are many different strategies you could use to try and do that.
I decided to build up a set of bezier paths to do it. Using several different techniques in combination to build up the complete set of feature paths. Using the bezier path layer to hold them as they are built up from different action steps in the PAseq.
You can see everything i used to do this in the PASeq shown above. And you can try the individual action steps out in Studio Artist by running the individual actions (press the red keyframe to run an individual action step), and look at the contents of the bezier path frame when in Bezier operation mode to see what that action step did.
Note that i first erase the bezier path layer, before i run commands to auto-generate paths. if i did not do that, then anything already inside of the current bezier path frame would contaminate my sketch.
I used 2 ip ops that can auto-generate bezier paths (Edge ip op, and Smart Contrast ip op). To use those to auto-generate bezier paths, you setup their controls in the image operation Editor control panel, and then run the Path : Generate Path : from Ip Op menu command.
If an ip op does not auto-generate bezier paths, that menu will be grayed out.
I used 2 different image operation effects to auto-generate bezier paths, as well as the Path : Generate Paths : From Source Edges menu command.
I also used various Path : Path Layer Commands menu commands to try and straighten out the bezier paths, and auto-connect ones that were close to each other, and also filter out really small paths.
After generating my feature sketch lines and storing them in the bezier path layer, i then used the Path : Paint Path : Full Layer menu command to automatically paint in the auto-generated bezier paths with my 1 pixel hard edges black paint preset.
A true WPAP style uses very straight lines for the sketch. So this part of my automatic sketch process is the weakest part of my attempt documented in the PASeq. If you are really trying to get a true WPAP look, the image feature sketch lines rom this approach are less than ideal. The global stylistic fracture patterning on the other hand i think matches the WPAP style quite nicely.
So i don't consider the finished output of this PASeq to really be representative of the WPAP style, but i thought the overall PASeq strategy it uses, and how it was put together would be useful information for people. Because it shows off a lot of different and useful techniques associated with bezier path layer manipulation you can use in your own work.
Another approach would be to use the paint synthesizer directly to build up a rough feature sketch using short straight line segments. I'll try to post an example of that kind of alternative approach later.
Replies
Thank you John.
My images look less clean than yours.
The edge detection algorithms that generate the bezier paths are picking up a lot of background noise that is in your particular source image and then generating extraneous small bezier paths associated with that noise or texture.
Preprocessing the source image you want to work with when running automatic effects is a way to try and deal with issues like this. You might want to selectively blur the background areas of your source image, and then use the preprocessed modified source image instead of the original one as the source for the automatic PASeq preset.
You could do this by using one of the selection tools to select the background, and then use something like the Smooth ip op with the Algorithm set to Blur Edge and then the selection masking i just described turned on when you run the ip op effect to only smooth out the background areas of the source image (while leaving the foreground feature areas of the source image untouched). Increase the size of the Spatial parameter if you want more of a blur effect when using that ip op.
Or, you could alternatively use a paint preset that blends or blurs the canvas to manually blur or blend out those noisy areas in the background of the original source image. If you have the old Toolbox paint preset collection, there is a category in there called Blur that has all kinds of useful blur and blend paint presets.
I actually did do something just like i described to the source image i used for my test above. And for my other WPAP tutorials.
First i dragged my original source image into the canvas.
I then manually selected the background using a lasso selection tool, and then ran the Smooth ip op with masking turned on.
I feathered that selection before using it to mask the background with the Smooth ip op effect. You can use the Canvas : Selection : Feather menu command to feather the current selection mask. Feather means the edges of the selection mask are softened.
I then used the Blend Blur Min paint preset in Toolbox : Blur to do some additional manual touch up to the background of my original source image.
I then inverted the selection, and ran a sharpening ip op effect to boost the edge detail in just the facial regions of that source image a little bit. I then saved that modified source image (so that i could use it later), and then opened the modified source image as my new source image for automatic processing.
I used that pre-processed Miranda Kerr source image for all of the experiments i have posted recently using that image.
There is also a Canvas : Canvas to : Source menu command you can use if you don't care about saving the pre-processed source image for later use. This menu command can also be recorded into a PASeq preset as a recorded action step. So you can build a single PASeq preset that pre-processes a source image automatically, then loads that into the source, and then uses the modified source for the rest of the PASeq processing.
So, my pre-processing strategy for my specific source image had 2 different components to it. I wanted to get rid of extraneous detail in the background areas of the image. And i wanted to actually boost the foreground facial detail a little bit so that the automatic algorithms i would be using would have enhanced facial features to work with. In the hope that would enable them to better automatically reproduce facial feature details.
A little bit of intelligently designed pre-processing of a source image you are planning to use with automatic painting or bezier curve generation presets can go a long way towards improving the results you can get from the automatic drawing methods.
There are also 3 different Size Prune menu commands in the Path : Path Layer Commands menus. They remove progressively larger small bezier paths as you move from Size Prune to Size Prune1 to Size Prune2. So that's another way to clean up small bezier paths generated by noise or unwanted texture features.
The tradeoff is that as you filter out small bezier paths, you might also be removing small details in facial features that you would really want to keep around. How that plays out could be different from image to image.
"There is also a Canvas : Canvas to : Source menu command"
Had I known this before I would have saved SO MUCH TIME and used SA for many more applications. Any chance there is a way of integrating this for movies? I find most animations need two rounds of processing one the animation style and the other a temporal imop blending which can only be done once the first animation is made and complete. Being able to automate that would be grand.
You can record that menu command as an action step in a PASeq you use for processing movies. So yes, you can use that as a part of your movie processing.
You're missing my use case. After rendering the rotoscoped film clip I'll do a pass with a Centered x5 temporal imop to the whole movie to smooth over flicker. Far more consistent results than overdrawing in my experience. So in that instance loading a single frame in a PASeq does not help whereas dropping the whole rendered clip and running a second PASeq does. It may just be the kind of thing that does not work to automate but wanted to mention.
Excellent John.Thank you.
This is a first class lesson about how to pre-process an image in SA.Mandatory.
WPAP. In search for clues.I obtained the image below using as a source, a B&W image colorized in the luminosity zones.Then I applied the IpOp preset Line Screen Regionize with the values I've found in the PASq Light Screen Regionize4.The patterns show clearly the limits of each luminosity zone and interestingly, they are variable.
One question arise without considering the technical requirements of SA,which I don't know: Is it possible to generate geometric shapes of the luminosity zones?.They should take the information of the real zones ,but could generate random shapes inside of the real ones.One restriction is that the limits between zone must be given for the differences in luminosity and not for solid lines,but I think this will not be a problem.
The colors of the different zones must be in accordance with the luminosity zones.
If you are using the vectorizer, then the Region Effect control setting (Drawing control panel) allows for various geometric stylistic effects to be applied to the individual regions.
And the Shape option for the vectorizer Main Technique is really all about generating geometric shapes,as opposed to regions that try to closely reproduce all of the subtle details of the original source image.
Line Screen Block Regionize ip op is all about filling in geometric block shapes.
There are also different controls in the paint synthesizer, like the Stylize Effect option in the Path Shape control panel that convert a path into different geometric stylized paths.
And if you are working with Region Fill as Brush pen mode then there are ways to do similar things, get a geometric stylized path for the region shape rather than a more complex organic shape that directly is derived from the natural contours in the source image.
There are also all kinds of MSG processors that generate geometric shapes. Some of those could be modulated with a source image, or some property of a source image. So that would be a totally different approach to generating geometric shapes.
And i discussed one approach to generating geometric shapes in the tutorial above by being clever about how you program the paint synthesizer to auto draw lines. So you could come up with all kinds of different strategies to do that, that would create different geometry than the centered fracture pattern i created with that approach in the tutorial above.