User:I am not Seedfeeder/Vectorising the Seedfeeder images

Let's take File:Wiki-cowgirl.png as an example (1).

Almost all the images contain only:

• one or more characters
• shadows
• a gradient background.

I figure you know how to do the gradient background, so we won't bother with that-- except to suggest you do it last, so that you can check the alignment is right.

In your bitmap editor

Here, I'm using GIMP. I'd expect other similar software to work fine.

Split out each character (2, 4) and the shadows (3). We'll come back to the shadows in a moment.

Once for each character

The next part should be done separately for each character. Let's use the person on top (4) as the example.

• Sketch in the parts which are hidden behind something else, like their right knee. Don't bother doing a lot of detail, because this won't be visible. But if you misalign something slightly, and you haven't drawn in these parts, the whole picture will look odd. Also, it helps with lining things up. (For example, on this particular picture, I drew in the rest of the penis of the person underneath, which clearly had to coincide with the other partner's vulva.)
• Make a silhouette layer of the character (5) for later use.
• All the Seedfeeder pictures are drawn with thin dark lines, then coloured in beneath. See if you can find all the lines, and produce a layer containing only those lines, in black (whatever their original colours were), and the rest transparent. We'll call this the outline layer (7).
• Lastly, figure out which parts of the colouring are skin tones. Parts which aren't skin tones often include
  • head hair
  • pubic hair
  • often genitals, especially vulvas. This wasn't an issue here, but compare File:Wiki-cunnilingus2.svg
  • eyes
  • nipples
  • makeup, though you might decide it's not worth the fuss
  • jewellery (finger rings, earrings...)
• Separate those out into separate layers, which we'll call detail layers. (However, eyes and nipples are so small that you can ignore them for now, and just put them in with the vector editor later, as circles.) Delete those areas from the skin layer.
• Now select all the lines in the outline layer. Replace those parts on the skin layer with skin colours.
• Next, take copies of each detail layer and the skin layer. You'll be needing to refer to them later.
• Back on the layers themselves, stretch the colour contrast of each one.
• Then posterise down to as few colours as you can get without sacrificing quality. (This is why you had to delete the detail areas from the skin: it would affect the number of colours needed.) You'll end up with something like 7.

...and the shadows

Posterise the shadows in the same way. You should also make the shadows slightly larger than they appear, to give you room to make slight mistakes.

In your vector editor

I'm using Inkscape, which edits SVG files natively.

Once for each character

For each character,

• Import the outline layer, and vectorise it (to a monochrome path, like 7).
• Import the silhouette layer, and vectorise it (to a filled monochrome path, like 5).
• Import each posterised detail layer, and the posterised skin layer. Vectorise each of these to however many colours you used in the posterise step for that layer. (You don't need to remember the number, because you can see in the preview: when you increase the number of colours in the trace, it will make no difference beyond the number of colours.) If you have the option, smooth, stack, and remove the background.
• Delete the imported bitmaps.
• You might be able to simplify some of the colour layers further without ruining the picture. If you can, do. But don't try this with the outline layer: it needs to be sharp.
• For each colour path in the colour layers, set the colour to the closest colour in the original layers before stretching contrast and posterising (you kept them, remember?)
• Put the outline layer and the silhouette layer into a layer group named after the character, like "person underneath".
• Then create another layer group within that layer group, called "paint" or something like that, and move all the colour layers in.
• Add a blur to the paint layer so that it doesn't look like a contour map. You might want to group detailed layers (like genitals) together and blur them less than the others.
• Set the silhouette layer as a clip to the paint layer, so the edges remain sharp.
• Make sure the outline layer is on top of the paint layer. Turn down the opacity to maybe 40%, so that it matches the way the lines are drawn in the original.

The overlap layers

The natural sequence of the SVG layers will take care of one partner's body obscuring the other's. But sometimes, as in this image, parts of both characters obscure parts of the other. Here, part of the left thigh of the person on top hides the right hip of the person underneath, but the other part of the thigh is hidden by the person underneath's right arm.

Where the character below obscures the character above (above in SVG layer terms, not above in the sex position), you need to create another layer group on top of all the others, and put those parts into it.

• The easy way to do this is also the most fraught: just redraw those parts of the body, as if they were an extra character by themselves. This causes some positioning weirdness and risks the colours not bleeding over the breaks.
• The better way is to make a new filled path which only covers the overlapping parts, and put that into the topmost layer group. Then clone the underneath person, and set your new path as a clip for that clone.

The background

• Import the shadows, and do much the same as you did for each character, including the blur. Don't forget to put them at the bottom :)
• Draw a linear gradient as the background, using colours sampled from the original image.

The result

It should look something like this:

Consider pulling that SVG apart and seeing how it all works.

Compare the original:

Afterword

• Suggestions and feedback (both for the images and for this document) are welcome, of course.
• Seedfeeder themselves planned to create a tutorial like this one, but only got as far as starting it: [1]