Character Creation HOWTO

Table of Contents



1. Introduction

I'd like to meet the guy who invented beer, and see what he's working on now.

1.1. Purpose of this Document

This document describes how to create additional characters for OpenMortal. The process is slow, painful, non-trivial and requires a lot of work, so if you only have one free hour, you might as well quit now. Then again, if you are dedicated and have the resources, you might find that creating a character, and seeing it work in the game is in fact fun and rewarding.

1.2. Intended Audience

This document is indended for people who want to create additional characters for OpenMortal. It only cover the creation if natural characters, e.g. not animated or rendered characters, but real people. You don't need to be skilled in martial arts, expert in programming (although it probably helps if you are both .. just kidding!).

1.3. Revision History

Revision
Date
Author
Description
1.
2003-09-28
UPi
Initial version. Describes the creation process as of version 0.3
2.
2003-10-01
UPi
Finished the gimping part, added overview.
3.
2003-10-03
UPi
Added sample gif.

1.4. Overview of the Process

  1. Get an idea of the character in your head.
  2. Set up the recording equipment, and record every move of the character twice. Result: 5-10 minutes of DV video footage.
  3. Copy the video to a computer. Result: ~2-4 GB uncut video file.
  4. Cut the video file. Result: 60-90 seconds of cut video.
  5. Export the cut video at 1/2 frame rate (12.5 frame per sec for PAL, 15 frame per sec for NTSC) as png. Result: 800-1200 png image files.
  6. Create the learning samples for Contour. Result: learn_original.*.png and learn_foreground.*.png
  7. Run contour. Result: 800-1200 contoured and cropped png image files and neural.log
  8. Import the files into The GIMP. Result: one xcf file with 800-1200 layers.
  9. Remove unnecessary layers, and do the labeling. Result: one xcf file with 280-350 layers.
  10. Fix the images, align and retouch them.
  11. [Final steps are done by the OpenMortal team: converting to a DAT file and coding]

1.5. Resources


2. Getting a Bright Idea

"Owl, I have an idea", said Pooh.
"Nice, fat, clever bear you are", said Owl.

This might seem trivial at first, but before you start recording, editing, processing, retouching, etc. a character, you have to create it in your mind. Think about the character a bit.

What does he look like? OK, obviously he will will look like you or someone you know... but does he think, act, move or behave the same?

What should the character wear? Some good outfit can greatly enhance the character. Are there any other weapons and tools that he uses?

Think about the characters nature and demeanor. Is ha a good guy or a bad guy? Or just plain confused? Can he actually do any martial arts, or is this his first time in a fight?

How is the character different from the other characters already in OpenMortal? Having a different backstory is cool, but is there something in the character that actually makes him unique?

Finally, how does he move? How does he do his kicks, punches, and other regular moves? What are the special moves of the character?

Once you know the answer to all these questions, you can move onto the next phase.

3. Shooting a Character

"He's dead, Jim."

For the people of Texas: it does not involve using a shotgun, and you should not attempt to involve one either.

Shooting is perhaps the most techinally difficult part of the character creation process. It's like always with content: garbage in - garbage out. If you mess up the recording (e.g. by insufficient lighting, bad camera settings or inadequate background) you will not be able to fix it in later stages, and  the character will be dead before he makes it to his first fight.

3.1. Setting up

The first time I tried creating OpenMortal was in 2000. Back then, I didn't have a digital camcorder, only a camera. So we tried and tried to create characters from a large number of digital still images. Then we tried some more. You can imagine how difficult it was to do the falls.. Here's an image of a failed attempt of recording the fall.. My soul still screams in terror when I remember those days.

Here you see several mistakes: The background is white, which meand that the shadows (which are of course a dark shade of gray) are hard to distinguish from the black parts of the character (the trunks and the hair). The person helping is not covered with the background color. The lighting is bad: the flash of the digital camera is reflecter right back at the camera, making everything look dark backlit. All in all, this attempt was quite terrible. I spent hours on retouching, re-editing, color balancing, etc. about a 150+ photos.  The results were disappointingly low quality despite my best efforts.

Fast forward to 2002: I have come a long way. I got a digital camcorder, proper lighting, some blue cloth for background. After some playing around. I replaced the blue background with a green one. It simply worked better. Here's how I set up the recording "studio":
[IMG: Recording setup]
The proper recording setup. This is how it looks from above.

The most important points are:
Setting up takes some time. Calculate at least one hour for setting up, and cleaning up after the recording.

3.2. Recording Checklist

First of all, don't forget to warm up and stretch. Some of the moves are difficult and you don't want to strain yourself.

Here is a list of the stuff that needs to be recorded:

3.3. More on Recording

It is a good idea to record many characters at the same time. It is more fun, you can draw ideas from and encourage each other. On the great recording session of 2002, we recorded 10 OpenMortal characters in one shot!

Tips for the director:
Tips for the actor:
Tips for the cameraman:
After the recording you will be left with a 5 to 10-minute DV footage per character. The size of this thing is horribly large, in the order of Gigabytes. The next steps ensure that it get smaller, and usable in the game.

3.4. Sample Video

Since recording is THE most important part of the character creation, I decided to include an example video here. It is 3.4MB in size, and I have compressed it horribly, so only download it if you want to learn the recording, because it will be useless for anything else.

4. DV Editing

"Most editors are failed writers--but so are most writers" - T.S. Eliot

IMG: Editing with KinoOK, so in the previous step you produced the footage. It's large, full of recorded mistakes, and interlaces - quite useless for the game. In this step you deinterlace the footage, and remove the useless parts, leaving a cut version, which should contain only the parts that you need.

4.1. Copying the Video To Your Computer

Linux users: Transfer the footage from the camera to the computer is usually painless. I used a program called dvgrab which works from the command line, and does the job just fine. See the documentation of dvgrab for details (e.g. kernel modules that need to be loaded..)

Windows users: Usually the firewire/video editing card comes with some software drivers and GUI programs that can grab from a DV camera. Consult your hardware documentation.

4.2. Cutting the Video

Most nonlinear video editors (such as MainActor or kino) can be used to splice the footage, and remove the useless parts. In the previous step you recorded every move twice, now is the time to see which one came out best, and remove the other. You may also remove the second part of a move if it is the same as the first part, played backwards. This saves some work later on.
After you're done editing, export the video into image files (e.g. png format) at half framerate. The half framerate for PAL cameras is 12.5 frames per second, for NTSC cameras it is 15. At this point you end up with somewhere around 1000 pictures. Don't forget to deinterlace the images, this should be a feature of any video editor.

5. Using Contour

"I DO NOT HAVE A GUT! I have... contours." - Frasier

OK, I'll start with the frankness. Contour sucks! It is hard to use, the results are crappy and it is slow as hell.

Trouble is, I don't know of a better tool.

Here's what you have to do to make Contour work.

5.1. Creating Learning Samples

The learning samples are, simply put, some pairs of images which are designed to show Contour what is the foreground and what is the background. Based on these learning samples, Contour will try to extrapolate the rest of the images by itself.

Each learning sample consists of two png images: one will be called learn_original.x.png, the other learn_foreground.x.png. In these filenames, x denotes an integer, starting from 0 up. The original files are simply cut from one of the png files which were produced in the previous phase. Only cut the important part of the file, do not include too much useless background, and nothing from the environment. The foreground file has to be the same dimensions as the original file, but it has to be black and white: black marks the background, white marks the foreground.

You need to create at least two such samples, which means at least four files:

learn_original.0.png
learn_foreground.0.png
learn_original.1.png
learn_foreground.1.png

All of these files should be placed in the same directory as the 1000 png files. Here are example learning samples:

learn_original.0.png learn_foreground.0.png
learn_original.0.png learn_foreground.0.png
learn_original.1.png
learn_foreground.1.png
learn_original.1.png learn_foreground.1.png

It is vital that the samples are of good quality. If you make the edges too rough, Contour will actually learn to include parts of the background or remove parts of the foreground! So strain your eyes, or adjust the gamma of your display.

Try to choose a typical and an atypical sample. If the learning doesn't work, you might need to choose some other images to learn from. It is usually bad to use more than two learning samples.

5.2. Teaching and Running Contour

So you have this directory with all the png files, the learning samples and everything! Great!

Now you need to fire up Contour. First open the directory which contains your files. Next, choose the "Create Neural Net" item from the menu. A dialog such as this one will appear:

[IMG: The learning dialog]

The learning process is extermely slow. It might take several minutes until the result is acceptable.

Next you have to fine-tune some parameters: these parameters are the treshold (from what level is the output considered "foreground"), and some blurring (used for smoothing the contour). Experiment to see what parameters work best with your images. Remember, once you start the "Run Directory", these parameters will be used for every image!

Once the learning is satisfactory, click on stop, and choose "Run Directory" from the menu. Here you need to choose a "source" directory (make it the same as the directory where the learning samples and the png's are), and a destination directory (create an empty directory for that). Click "Start". Once this is started it cannot be stopped!

[IMG: Contour is running.]

The neural network will run for every image in the source directory, and produce a cropped output image in the destination directory.

6. Using The GIMP

"The Gimp is a very complicated program for users and developers alike" - First sentence of the GIMP documentation.

The GIMP is the "GNU Image Manipulation Program", a free, extendable and scriptable image editor (think "PhotoShop").

The objective of this phase is to further decrease the number of images, and to "fix" those images that were messed up by Contour.

In order to complete all the steps that are required here, you will need a copy of The GIMP, and also the Perl extensions to the GIMP, also known as Gimp-Perl. The GIMP itself is available for Windows and Linux (as well as many other UNIX-like operating systems), however the Gimp-Perl is only available for Linux. This will be a problem for Windows users in step 6.1.

The GIMP can be downloaded from http://gimp.org/download.html for Linux, or http://www2.arnes.si/~sopjsimo/gimp/#download for Windows. The GIMP and Gimp-Perl are now part of most recent Linux distributions, so you can probably install it from your CD's or get the package from your distribution's local mirror. For tutorials about The GIMP, see this page.

Before you continue:

6.1. Using The GIMP

Although introductions and tutorials for The GIMP are abundant, and some of them are quite well-written, I will nonetheless venture to write a short guide here. I will specialize on GIMP functions that are the most useful for creators of OpenMortal characters.

Loading and saving. The GIMP's native file format is called xcf. While you work, you will need to save your image as an xcf file. Saving in other formats will cause the loss of information (e.g. layers). Under linux, you can save the file compressed by specifying an .xcf.bz2 or an .xcf.gz extension. Compression can take a while on slower computers, but can save some disk space. (Actually I am told that you can save as .xcf.gz on Windows if you have a program called Minigzip installed) Although recent version of The GIMP are quite stable, you are encourage do save your work often.

Keyboard accelerators. Most operations in the gimp are invoked by pressing the right mouse button inside the image window and thus invoking the image menu. However,you can assign a keystroke to any menu item, by hovering the mouse over a menu item, and pressing the keys. Keystrokes can be single keys (e.g. A, or F1), or a combination (Ctrl+Shift+A). It saves a lot of time to customize keystrokes to suit your needs. The combinations that I use most often are:
[IMG: Layers in the GIMP] Working with layers. Layers provide much of The GIMP's power and flexibility. They can be combined in many ways to create a composite image. The Layers window can be opened by pressing Ctrl+L in the image window. One thing that is not so intuitive about working with layers is that the visible layer is not always the active layer.

6.2. Importing the png Images Into The GIMP

So, you made it this far. You have the GIMP, you have Gimp-Perl (or maybe you don't - Windows users read below), and you have ran Contour for all your 1000 png files. Excellent!

Contour has created, along with the output png files, a data file called contour.log. I have written a small Gimp-Perl script which will read this log, read the image files, and put them in one huge GIMP image. This script can be downloaded from here. The script must be copied into your gimp plugin directory (this is set when you install the gimp).

Install the script (you will need to restart The GIMP if it was already running), and start it from "Xtns/Perl-Fu/Mortal load". Enter the path to the directory where the contoured images and contour.log reside. After a lot of waiting, a new window will appear. It will be empty. Don't panic! If you open the Layers dialog, you will see that it holds all your precious images, each in it's own layer, but none of them are visible by default.

6.3. Removing Unnecessary Layers and Labeling Moves

First thing to do once you have importad the image is to save it. This will take a while, given the size and number of layers.

Once this is done, you can start the first and most important job: removing unnecessary layers and labeling them correctly. This is important for many reasons: it will save disk space, memory and download time for gamers, it will save you time (you won't have to fix and retouch so many layers), and it will make integrating the character into the game easier.

Once you are done removing these unwanted frames, you should be down from the initial 1000-1300 frames to around 300-350.

Here is a table which shows the number of layers each current OpenMortal character uses. When creating new characters, try to make the number of layers similar to these values. In particular, kneeling (referred to as "crouching" earlier in this document) should always be 4 layers, because it is used by the jumping algorithm. If a move of your characters is particularly simple, make it fewer layers, if it is long and overdone, you can use more.

  Zoli UPi Cumi Sirpi Ulmar Maci Bence Descant Grizli
start 21 18 18 20 30 23 23 13 14
stand 4 5 4 7 6 5 4 4 6
walk 13 14 14 13 12 14 12 10 15
turn 9 8 8 9 8 9 8 9 9
falling 11 17 12 14 14 12 18 16 15
laying 1 1 1 1 1 1 1 1 1
getup 14 16 19 16 20 18 12 15 16
hurt 16 12 9 13 13 13 8 7 9
swept 6 6 6 13 9 8 8 10 16
won 11 19 26 10 14 17 8 20 20
block 5 5 4 5 6 5 5 5 6
kneeling 4 4 4 4 4 4 4 4 4
onknees 1 4 1 4 4 4 4 4 5
kneelingpunch 7 5 4 5 6 5 5 5 5
kneelingkick 4 5 5 6 5 6 - 5 5
fun 7 5 7 8 7 9 5 13 9
threat 9 6 6 9 5 8 6 7 12
highpunch 5 4 4 5 11 5 6 5 9
lowpunch 6 6 5 6 10 6 5 6 6
highkick 8 6 6 7 6 7 7 7 6
lowkick 6 5 6 6 6 6 6 6 6
groinkick 6 6 6 7 7 10 6 6 7
kneekick 6 7 6 7 7 7 6 6 8
elbow 5 7 5 6 7 6 7 5 12
spin 11 10 - 14 11 11 10 - 17
grenade 15 18 7 9 10 10 11 10 10
uppercut 9 8 8 8 9 8 7 7 17
throw 16 15 24 22 34 15 22 23 19
highpunched 9 6 5 5 6 8 5 5 11
lowpunched 4 5 5 5 7 6 6 6 6
groinkicked 13 19 17 9 8 14 8 9 8
kneelingpunched 4 4 4 4 5 8 4 5 5
kneelingkicked 3 7 5 6 6 6 5 5 4
thrown 4 13 7 14 10 12 8 8 14
specpunch 13 - 27 13 - - 14 19 6
funb - 3 - 12 - - - 11 -
sweep - 7 14 9 10 8 7 11 15
cranekick - 9 - - - - - - -
burninghands - 13 - - - - - - -
kneelingfire - 12 - - - - - - -
spit - - 5 - - - - - -
junk - - - - 1 - - - -
junkb - - - - 1 - - - -
specsomething - - - - 9 - - - -
scissorkick - - - - - 15 - - -
pierce - - - - - - 8 - -
knife - - - - - - - 17 -
hipshot - - - - - - - 9 -
nunchaku - - - - - - - - 15
stomp - - - - - - - - 11
Total 286 340 314 331 345 329 289 334 379

While you are removing the unwanted layers, label the layers. This means that for the first layer of every move, rename it to match the name of the move. You can see on the screenshot at the beginning of this chapter, how this was done for STAND and WALK in Grizli's case. Please use the names in this table (make up names for unique special moves).

6.4. Fixing Bad Contours

Every now and then, Contour will mess up. It might be that the picture is too dark or blurred in certain places, or something else that messes up the detection, but it will happen, and will happen a lot. To fix the mistakes of contour, you can use these tools:
Fixing and cleaning up 300-or-so frames is time consuming, but the result will be much nicer. Even if you don't use the eraser to fix the frames, you must fix the degenerate frames.

6.5. Aligning The Images

This simply means that the layers should be shifted in a way that the character will start every move from the same origo. Every move should be aligned to the start move (or the onknees move in case of the kneeling punch and kneeling kick).

To move a group of layers together, you need to click just right of the "eye" icon on every layer. Another icon (four arrows) will appear, and this means that the layer is now grouped. If you move any layer with the move tool (M key), the grouped layers will move with it. So, if you group every layer of a move, you can move and align them together (layers within the move will be properly aligned by Contour).

The best way to align the images is to
For example, in this image, the "HIGH KICK" is not aligned with the standing position (shown in the background.) The images which belong to the high kick are already grouped together (observe the arrows icon next to the eye icons), and ready to be moved to the left, and a bit down.

The following moves do not need to start at the START position:

6.6. Adding Special Effects

The GIMP offers a huge array of filters and other transformations that you can apply to the poor character. Here are some ideas:
The list goes on... Before you add any special effects, make sure to make a copy of the original! It is not nevessary to add special effects to a character to make him interesting. Since OpenMortal features real people, the idea is to make each character unique and original by making their demeanor different from the rest.

Also be aware of the "Filter all layers" menu at the top of the Filters menu... You can check the animation with the Filter/Animation/Playback menu item.

6.7. Saving the Result

Once you are done with all this stuff, save the xcf file. The rest of the steps should not be saved into the xcf file as they are not reversable and result in some loss of the original content. If you need to change the character later, you will need the xcf file intact, and replay the following steps:
You are DONE!! The result is a gif file which is about 3MB in size. You can watch it by opening it in a web browser.

Here is an example of what the result should look like.

7. Where Do You Go From Here?

You've Come A Long Way, Baby - Fatboy Slim

So you have successfully created a character's animation! And you only had to work on it for hours! So what now?

Answer is, the rest of the process is even more complicated and less intuitive than it has been so far. It involves using the MortalEditor to mark the head, torso and legs of each frame (a semi-automated process), check the hits, and create some code which makes the character work inside the game. If you want to do it yourself - fine, there are enough examples (the rest of the characters) to learn how to do it. But you shouldn't have to.

For now, your work is finished. You need to send me the gif file, along with a description of the character (name, age, etc., and what his special moves are supposed to work like). I will add the character to the game, and return it to you, ready to be played. To send me the character, please go to the Character Upload Page.

If you have questions or comments regarding this documentation or the character creation procedure, please feel invited to ask for help at the OpenMortal Help Forum.