Making Of 'The Gates of Hell'
The Gates of Hell was created for the 38th 3Dluvr contest entitled "the gates", which got 3rd place. I was wondering how many people in the contest would make the gates of hell and how would be the most obvious representation. I remembered Rodin's the gates of hell. I wanted to make the gates of hell but not the gates of hell of the hell itself, I wanted one gate sculpted in stone, but still very scary. Something high detailed, like baroque ornaments.
One of the first concept I drew the umbral with a lot of serpents
(the serpents has a lot of symbology in mythology and religions), then with confused details of rock and mystic ornaments and finally with a lot of nude human bodies representing their souls trying to move up to the paradise (or get out the hell). Mixing the concepts and re-elaborating some parts, I got the whole gate concept finished.
The planning is generally an underestimated process of execution, but in the gates of hell I had to take care in a special way as it was cleary very complex to manage (the final concept had the umbral filled with human figures, and a human figure requires a mininum number os polys to be recognized)
The first main problem I had was about human figures and how to manage the polycount and still position the figures, arrange elements and the realtime viewport continues at an acceptable frame rate. The other problem was the time, I had the contest deadline to meet. So I decided to start from the most complex part, I started modelling the human figure and use the layer system to manage. Is crucial to solve the problemas when working with complex projects.
I started positioning the camera and creating the wall and the floors to sketch the composition how it will be in the end to start adding details around it.
I started building the human figure at mid-poly, I modelled the full body then added the hands feets and the head. The head doesn't have eyeballs, it would increase the polycount too much when considering the model's mutiplying to fill the gate umbral.
Then after the bone configuration and character setup I made about 12 different poses from the same figure (the contest deadline and my spare time didn't allowed me to model than 1 human figure for this purpose). The character deformation is not 100% accurate but for this case it will serve considering the small size of each character in the final scene.
The Serpents head was made using box modelling technique like the human body and the body of the serpent was using spline and loft as shown in fig5e. In this case used helix for the path and circle for the shape.
Then the umbral was made using spline corner and then extruded to get the shape desired to work on. The umbral would not be seen too much, so the preliminary test with lightracer was made with a coupleof the human figures available to see how it was going to be.
The image shows the umbral test now more smoothed because I wanted to improve the details at the same time that I was not so sure about how much of the umbral would be seen on the rendering as far as the human figure should cover almost the whole umbral.
Then I tested the other side a little bit as well as new details on the top made with cut, extrude, welding, etc. Even in the tests, I separated the human figures (high polycount) in layers, because I could easily turn on and off and make the viewport lighter while I navigade and adjust the details, then I went back to camera, and turn on again; it's a pretty nice way to deal with a lot of polygons.
This image shows the umbral now with a very high polycount optimized because (as I will show during texturing process) is not so easy to get correct bumpiness when using lightracer. In this stage you can see both the umbral and serpent body and details with displacement.
Then was time to improve the details and make more tests to see how the serpents would integrate the umbral and human figures and the added details on the top, and the big and scary devil's head. the fig.5l shows the preliminary version of the devil's head, surely needed more details.
Focusing on the upper part, I turned off the side parts and rendered the now improved head with a lot of details on the umbral and human figures positioned and with the first attempt on making the door base
The next picture shows some tests on the wall (chamfer edge and extruding down a bit) if some modelling was needed, a little render with the wall and the door and decided to do not use this on the final version.
Then I started the door. The serpent head is the same, with some modifications as well as the door's serpent made with spline+loft the same process of the serpent of the umbral. the circle detail was made with a very thin torus, inside another one, the little balls are half hemisphere of geosphere and the triangle as well as upper details are spline converted to mesh with appropriated tickness.
The main problem on texturing process I encountered was the bump while using lightracer. If you use lightracer you know what I mean, if don't take a look below.
A standard bump map with lightracer usually get this result:
Viewport default displacement:
rendered version of default displacement on the mesh:
Ok, Mixing both results is possible to achieve better results at the final render.
Now the texture preparation:
There is no secrect trick on preparation of the textures, you can easyly find some very good wall/concrete textures, buy some of the great 3DTotal textures or take some with your own digital camera. What will make the difference is the care when dealing with details and customization (like the door texture). As far as almost every project overview here inside 3DTotal covers nicely texture creation and preparation through layer blending, I'll show only few steps I used to mix these one. In many case you'll note that you should make a couple experiences with blending mode to find the desired result, according the color and darkness of the layers involved.
Ok, sometimes you can't see too much difference between some blending modes, it depends on the color involved and when dealing with gray tones you may find a bit hard to achieve color blending mode, my advice is to make several test with the blending modes on Photoshop until you get confortable, so when you need them you will find the effect you want more easily.
For the door tiles I used this shape a few times and then erased a little bit the down part to fade and get a more gradual mix
and then blend to the concret texture I prepared before
The final (diffuse)textures for the scene are:
The snake body I used a similar tile
and mapped to a shape near the body of the snake and rendered it. Then positioned the figure over the mapped region as show figure below and changed the blend mode to overlay.
The Fleur de Lis wasn't so difficult at all, Photoshop has the shape on it's default library, just access them at
Rasterize the shape and then blend it to the concrete texture below
When settting up UVs for highly detailed things, Sometimes is necessary adopting different strategies. If it's a flat color, no UV is needed at all, in this case, I wanted a concrete/stone look (UV requiered), but if I Unwrapped each human figure as I would if it's highres, I would have much more time spent and probably would not reach the deadline at safe time. A good solution I 've found is collapsing a good couple of human figure on the umbral, and then applying a single UV mapping to them all and then repeating the process until every single one is grouped, collapsed and mapped. For the walls a planar mapping will be fine (exept the holes dettached and maped individually.
I used a single sky light with lightracer with the following customization parameter:
On the side viewport you can see the scene setup
Now the configuration of Lightracer
you can click 9 or go to menu Rendering> Advanced Lighting.
Important parameters to change:
Bounces values determines the bouncing values of the photon before they die, higher value, more bounce and more light (and higher rendering values).
the Color Bleed, will define the reflection of the object light
The adaptive Undersampling has to do with the final quality of the image, you can see more clearly when rendering GI for flat-color-object and you can see the flick when using higher Initial Sample Spacing. The rule is: For better quality and less flick, use lower initial sample spacing (like 4x4 or 1x1) with subdivide down to 1x1 however it will increase a lot the rendering time, you can balance these values and define which one fits better your needs.
note: You can check out my tutorial on Introduction to Lightracer on 3DCafe directly by clicking here.
After the imaged is rendered I added some brightness/contrast adjustment and a bit of color balance to Cyan and a bit of blue. I really recommend some post work to improve the lightracer render
The final result is:
hope you find this project overview useful
Mario Russo - 2004