To generate a realistic candle flame, the first thing to do, of course, is to light a candle and stare at it. What should become evident are some of the behavioral and visual elements which define its appearance. In terms of Maya lingo, a flame looks more like a transparent, incandescent surface than a nebulous object made out of thousands of particles. Flames have feathered yet hard edges, a very smoothly gradated interior, varying transparency and fluid behavior more akin to liquid or fabric than gas.

Based on these observations, a technique which will work well for single flames viewed from close proximity is softbody geometry. This technique will involve two phases: rigging how the flame behaves and how it renders. Again, our focus is not on how to create fire, as one would see in a fireplace/campfire/fireball. Those effects would lend themselves more to particles than geometry based on Maya's engine. Ideally we would like to combine both techniques, to create iso-surfaces from particle masses, a technique which may become viable via Fluid Dynamics in Maya 4.5. Studios such as PDI have written propreitary software for this, used for the Dragon's fireballs in the film 'Shrek', but we'll have to wait.

This tutorial will make several assumptions about your knowledge of the software as I am going to try to keep this as short as possible. We will be addressing modeling, softbodies, goals, springs, fields, lights, hypershade, render utilities, the connection editor and shader glow.

1) model a flame using any technique you choose. The main thing is to be aware of the UV coordinate space of the flame, and if it is a poly surface, layout the UVs cleanly so that color ramps can be smoothly applied across the surface. For my example I created a NURBS curve and revolved it. Since we will be turning it into a softbody, make sure there are enough vertices to be able to get smooth dynamic deformations.

2) turn the flame into a softbody and design its behavior. Select the flame, open the 'Create SoftBody' dialog window and choose the following options: Duplicate Make Copy Soft, Hide Original, Make non-soft a goal, and set the Weight value to '1'. At this point the softbody particles will not move at all, unless the goal moves, because their goalPP values are at 1 and the goalWeight value is at 1

3) Use the Component Editor to set goalPP values so that the particles which surround the wick of the candle have a high goalPP, such as .8, but the rest of the particles should have a very low goalPP, such as .2

4) Add a turbulence field to the particles with a low frequency and animated phase. At this point the flame will have a 'tear' between the rows of particles where the goalPP changes from .8 to .2

5) Add 'springs' to the softbody to give the flame a fluid, clothlike behavior. In the creation dialog choose 'wireframe' with a value of 2. You will now begin experimenting with the turbulence field's magnitude, frequency, phase as well as the spring's stiffness and damping. As with any use of springs, if your springs 'explode' due to high attribute values, remember to increase the scene's 'oversampling'.

Once finished with the flame's dynamic animation, it is now time to focus on how it renders. This will require us to pay special attention to the brightness, coloring, transparency and edge quality of a flame. A flame's color tends to be bluish at the base but orangish towards the tip... sometimes reddish. This can be easily achieved by mapping the surfaces color with a ramp. A flame also tends to be brighest in the center. This can be mimicked by turning on Translucency for the material. It is the transparency which will be a bit more involved. A candle flame is more transparent in the center than the edges, and more-so at the base.

First apply the color ramp to the material color, raise the translucency value, and place a point light with appropriate color, intensity and decay in the center of the softbody. You will want the point light to move around with the flame, so connect the WorldCentroid attribute of the particleShape node to the Translate vector of the Point light. If your point light needs to cast shadows, remember to turn off 'casts shadows' on your flame geometry.

Now on to transparency.

Above we are looking at six flame surfaces each with a different shader which is controlling the transparency in a different way. Each shader is Lambert with a white surface color. Refraction is off.

A) no transparency
B) Transparency set to 50% gray. This is how Maya calculates transparency by default. As the transparency approaches white, the surface fades evenly.
C) Transparency directly controlled by the Facing Ratio of a Sampler Info Utility
D) Transparency controlled by a Set Driven Key relationship where the Facing Ratio of a Sampler Info Utility is the driver.
E) Transparency directly controlled by a ramp.
F) Transparency controlled by a multiplication of a ramp by the Set Driven Key curves generated by a relationship where the Facing Ratio of a Sampler Info Utility is the driver. Could also be achieved by connecting the Facing Ratio curves to the Color Gain of a ramp which is directly connected to the transparency.

While Render Utilities can seem mysterious due to their lack of documentation, they can serve a wide range of purposes. In this case we are relying primarily on the 'Sampler Info Utility'... not to be confused with the similarly named but completely different 'Particle Sampler Utility'.

The Sampler Info Utility (SIU) has several attributes, but one of its 'float' attributes is called 'Facing Ratio'. This attribute has a value between 0 and 1 which is determined by the ratio of normals on a surface to the position of the camera. When the normals are pointing to the camera, a value of 1 is returned. When the normals are perpendicular to the line of sight of the camera, the value is 0. By using the Connection Editor to connect the 'float' Facing Ratio Output of the SIU to the 'vector' transparency of a material, we will get a linear relationship between transparency and the facing ratio.

Above we see the results in Hypershade of making that connection. While the results are clear on Surface C, the drawback is the lack of flexibilty. What if you want the opposite effect, where the edges become more transparent, or what if you want a more dramatic falloff like Surface D.

Above we have created a Set Driven Key relationship where the Facing Ratio is the driver and the transparency R,G & B are the driven. This way, we can use curves in the graph editor to control how the falloff is achieved. The trick to getting this to work is that once you have loaded the driver and the driven into the Set Driven Key (SDK) window, and have keyed them to create the relationship curves, you will not be able to set a second key. This is because you can not manually modify the driver (facing ratio). Therefore, you end up with 3 curves in the graph editor when you select the material node, but each curve (R,G,B) will only have one keyframe at 0,0. At this point you will need to manually add keys to each curve so that each curve has at least two keyframes... one at 0,0 and one at 1,1. Once you have this figured out, you will want to turn on weighted tangents, unlock the tangent weights of the keyframes, and modify the tangents to control your transparency falloff based on the facing ratio driver. This will allow you to achieve results such as surface 'D' above.

The last step is to address the fact that not only are a flame's edges more opaque than the center, but the base of a flame is more transparent than the top. To achieve this, once the Set Driven Key/SIU relationship has been made, disconnect the outputs of the SDK curves from the material. Then create a ramp and a Multiply/Divide Utility. Connect the ramp's 'color output' to the Multiply node's 'input one', and connect the SDK curves to 'input two'. Then connect the output of the Multiply node to the materials transparency. If you wish, you can avoid using the Multiply node if you connect the SDK curves to the 'color gain' of the ramp node.

The final step in this process will be to apply Shader Glow. Above we see the progressive results of adding glow. The left-most image is without glow, but our SIU based transparency is evident. Once glow is enabled, our geometry remains, but this is optional. By turning on 'Hide Source', the actual geometry will not render; just the glow. This is a great technique which works well for the candle flame, but is not limited to it, of course. Even ghosts can be successfully achieved using a similar workflow to what we have been doing.

When glow is enabled, there are some important things to know about the Shader Glow node. In the multilister/hypershader is a Shader called Shader Glow which controls the look of -all- materials which have glow enabled. If you need different glow looks for different objects in your scene, you will need to render in passes. Furthermore, there are many attributes which can be tweaked to design the look of the glow, one of which is called Threshold. In the above image, the 3 right images have Threhold values of 0, .08, and .04. Threshold clips the luminance values which get glow applied and is clearly a very useful attribute to help design the desired look. Another important attribute to change is Auto Exposure which is on by default. This can cause glow to flicker in an animation, so it is best turned off. Doing this may cause your glow to get blown out, at which point you simply need to lower the glow & halo intensities on the Shader Glow node.

Another issue to raise is that we have chosen to create the candle by mapping the flame surface's color, not its incandecance. The is because we are relying on translucency to brighten the flame. If you need to use incandecance, remember that areas which are transparent will render brightly if that region has a high incandecance value. Therefore you would also want to use the SIU to map the incandecence, but with the reverse values.

Above we have the graph of a shader which is using incandecence to determine the brightness of the flame. The same SIU SDK curves are being used for both transparency and incandecence. But for the incandecence, the SDK curves are first going through a Reverse utility.

So that about wraps it up... If you are unfamiliar with any of the tools used, it is a good idea to get comfortable with them individually, before trying to recreate my results immediately. The more unknown things you throw together, the harder it will be to problem solve.