Copyright 2006-2008 by Albert Cahalan for the Tux Paint project
New Breed Software
March 8, 2006 - July 7, 2008
This HOWTO assumes that you want to make an excellent Tux Paint stamp, in PNG bitmapped format, from a JPEG image (e.g., a digital photograph). There are easier and faster methods that produce lower quality.
This HOWTO assumes you are dealing with normal opaque objects. Dealing with semi-transparent objects (fire, moving fan blade, kid's baloon) or light-giving objects (fire, lightbulb, sun) is best done with custom software. Images with perfect solid-color backgrounds are also best done with custom software, but are not troublesome to do as follows.
If you wish to submit artwork to the Tux Paint developers for consideration for inclusion in the official project, or if you wish to release your own copy of Tux Paint, bundled with your own graphics, you need an image that is compatible with the GNU General Public License used by Tux Paint.
Images produced by the US government are Public Domain, but be aware that the US government sometimes uses other images on the web. Google image queries including eitherYour own images can be placed in the Public Domain by declaring it so. (Hire a lawyer if you feel the need for legal advice.)
site:milwill supply many suitable images. (Note: the *.mil sites include non-military content, too!)
For personal use, any image you can legitimately modify and use for your own personal use should be fine.
Image Size and Orientation:
You need an image that has a useful orientation. Perspective is an enemy. Images that show an object from the corner are difficult to fit into a nice drawing. As a general rule, telephoto side views are the best. The impossible ideal is that, for example, two wheels of a car are perfectly hidden behind the other two.
Rotating an image can make it blurry, especially if you only rotate by a few degrees. Images that don't need rotation are best, images that need lots of rotation (30 to 60 degrees) are next best, and images that need just a few degrees are worst. Rotation will also make an image darker because most image editing software is very bad about gamma handling. (Rotation is only legitimate for gamma=1.0 images.)
Very large images are more forgiving of mistakes, and thus easier to work with. Choose an image with an object that is over 1000 pixels across if you can. You can shrink this later to hide your mistakes.
Be sure that the image is not too grainy, dim, or washed out.
Pay attention to feet and wheels. If they are buried in something, you will need to draw new ones. If only one is buried, you might be able to copy the other one as a replacement.
First of all, be sure to avoid re-saving the image as a JPEG. This causes quality loss. There is a special tool called jpegtran that lets you crop an image without the normal quality loss. If you want a GUI for it, use ljcrop. Otherwise, use it like this:
jpegtran -trim -copy none -crop 512x1728+160+128 < src.jpg > cropped.jpg
Bring that image up in your image editor. If you didn't crop it yet, you may find that your image editor is very slow. Rotate and crop the image as needed. Save the image — choose whatever native format supports layers, masks, alpha, etc. GIMP users should choose "XCF", and Adobe Photoshop users should choose "PSD", for example.
If you have rotated or cropped the image in your image editor, flatten it now. You need to have just one RGB layer without mask or alpha.
Open the layers dialog box. Replicate the one layer several times. From top to bottom you will need something like this:
- unmodified image (write-protect this if you can)
- an image you will modify — the "WIP" layer
- solid green (write-protect this if you can)
- solid magenta (write-protect this if you can)
- unmodified image (write-protect this if you can)
Give the WIP layer a rough initial mask. You might start with a selection, or by using the grayscale value of the WIP layer. You might invert the mask.
Warning: once you have the mask, you may not rotate or scale the image normally. This would cause data loss. You will be given special scaling instructions later.
Get used to doing Ctrl-click and Alt-click on the thumbnail images in the layers dialog. You will need this to control what you are looking at and what you are editing. Sometimes you will be editing things you can't see. For example, you might edit the mask of the WIP layer while looking at the unmodified image. Pay attention so you don't screw up. Always verify that you are editing the right thing.
Set an unmodified image as what you will view (the top one is easiest). Set the WIP mask as what you will edit. At some point, perhaps not immediately, you should magnify the image to about 400% (each pixel of the image is seen and edited as a 4x4 block of pixels on your screen).
Select parts of the image that need to be 100% opaque or 0% opaque. If you can select the object or background somewhat accurately by color, do so. As needed to avoid selecting any pixels that should be partially opaque (generally at the edge of the object) you should grow, shrink, and invert the selection.
Fill the 100% opaque areas with white, and the 0% opaque areas with black. This is most easily done by drag-and-drop from the foreground/background color indicator. You should not see anything happen, because you are viewing the unmodified image layer while editing the mask of the WIP layer. Large changes might be noticable in the thumbnail.
Now you must be zoomed in.
Check your work. Hide the top unmodified image layer. Display just the mask, which should be a white object on a black background (probably with unedited grey at the edge). Now display the WIP layer normally, so that the mask is active. This should show your object over top of the next highest enabled layer, which should be green or magenta as needed for maximum contrast. You might wish to flip back and forth between those backgrounds by repeatedly clicking to enable/disable the green layer. Fix any obvious and easy problems by editing the mask while viewing the mask.
Go back to viewing the top unmodified layer while editing the WIP mask. Set your drawing tool the paintbrush. For the brush, choose a small fuzzy circle. The 5x5 size is good for most uses.
With a steady hand, trace around the image. Use black around the outside, and white around the inside. Avoid making more than one pass without switching colors (and thus sides).
Flip views a bit, checking to see that the mask is working well. When the WIP layer is composited over the green or magenta, you should see a tiny bit of the original background as an ugly fringe around the edge. If this fringe is missing, then you made the object mask too small. The fringe consists of pixels that are neither 100% object nor 0% object. For them, the mask should be neither 100% nor 0%. The fringe gets removed soon.
View and edit the mask. Select by color, choosing either black or white. Most likely you will see unselected specks that are not quite the expected color. Invert the selection, then paint these away using the pencil tool. Do this operation for both white and black.
Still viewing the mask, select by color. Choose black. Shrink the selection by several pixels, being sure to NOT shrink from the edges of the mask (the shrink helps you avoid and recover from mistakes).
Now disable the mask. View and edit the unmasked WIP layer. Using the color picker tool, choose a color that is average for the object. Drag-and-drop this color into the selection, thus removing most of the non-object pixels.
This solid color will compress well and will help prevent ugly color fringes when Tux Paint scales the image down. If the edge of the object has multiple colors that are very different, you should split up your selection so that you can color the nearby background to be similar.
Now you will paint away the existing edge fringe. Be sure that you are editing and viewing the WIP image. Frequent layer visibility changes will help you to see what you are doing. You are likely to use all of:
- composited over green (mask enabled)
- composited over magenta (mask enabled)
- original (the top or bottom layer)
- composited over the original (mask enabled)
- raw WIP layer (mask DISABLED)
To reduce accidents, you may wish to select only those pixels that are not grey in the mask. (Select by color from the mask, choose black, add mode, choose white, invert. Alternately: Select all, select by color from the mask, subtract mode, choose black, choose white.) If you do this, you'll probably want to expand the selection a bit and/or hide the "crawling ants" line that marks the selection.
Use the clone tool and the brush tool. Vary the opacity as needed. Use small round brushes mostly, perhaps 3x3 or 5x5, fuzzy or not. (It is generally nice to pair up fuzzy brushes with 100% opacity and non-fuzzy brushes with about 70% opacity.) Unusual drawing modes can be helpful with semi-transparent objects.
The goal is to remove the edge fringe, both inside and outside of the object. The inside fringe, visible when the object is composited over magenta or green, must be removed for obvious reasons. The outside fringe must also be removed because it will become visible when the image is scaled down. As an example, consider a 2x2 region of pixels at the edge of a sharp-edged object. The left half is black and 0% opaque. The right half is white and 100% opaque. That is, we have a white object on a black background. When Tux Paint scales this to 50% (a 1x1 pixel area), the result will be a grey 50% opaque pixel. The correct result would be a white 50% opaque pixel. To get this result, we would paint away the black pixels. They matter, despite being 0% opaque.
Tux Paint can scale images down by a very large factor, so it is important to extend the edge of your object outward by a great deal. Right at the edge of your object, you should be very accurate about this. As you go outward away from the object, you can get a bit sloppy. It is reasonable to paint outward by a dozen pixels or more. The farther you go, the more Tux Paint can scale down without creating ugly color fringes. For areas that are more than a few pixels away from the object edge, you should use the pencil tool (or sloppy select with drag-and-drop color) to ensure that the result will compress well.
It is very easy to ruin your hard work. Image editors can silently destroy pixels in 0% opaque areas. The conditions under which this happens may vary from version to version. If you are very trusting, you can try saving your image directly as a PNG. Be sure to read it back in again to verify that the 0% opaque areas didn't turn black or white, which would create fringes when Tux Paint scales the image down. If you need to scale your image to save space (and hide your mistakes), you are almost certain to destroy all the 0% opaque areas. So here is a better way...
A Safer Way to Save:
Drag the mask from the layers dialog to the unused portion of the toolbar (right after the last drawing tool). This will create a new image consisting of one layer that contains the mask data. Scale this as desired, remembering the settings you use. Often you should start with an image that is about 700 to 1500 pixels across, and end up with one that is 300 to 400.
Save the mask image as a NetPBM portable greymap ("
.pgm") file. (If you are using an old release of The GIMP, you might need to convert the image to greyscale before you can save it.) Choose the more compact "RAW PGM" format. (The second character of the file should be the ASCII digit "5", hex byte 0x35.)
You may close the mask image.
Going back to the multi-layer image, now select the WIP layer. As you did with the mask, drag this from the layers dialog to the toolbar. You should get a single-layer image of your WIP data. If the mask came along too, get rid of it. You should be seeing the object and the painted-away surroundings, without any mask thumbnail in the layers dialog. If you scaled the mask, then scale this image in exactly the same way. Save this image as a NetPBM portable pixmap ("
.ppm") file. (Note: ppm, not pgm.) (If you choose the RAW PPM format, the second byte of the file should be the ASCII digit "6", hex byte 0x36.)
Now you need to merge the two files into one. Do that with the pnmtopng command, like this:
pnmtopng -force -compression 9 -alpha mask.pgm fg.ppm > final-stamp.png