Yes, Modul8 uses bilinear on all the medias rendered in the composition. Basically bilinear creates a gradient inside the pixel using the colors of the adjacents pixels, using a simple linear interpolation. It is "bi" linear, because it is achieved on both axis.
This technic appeared in real-time imaging with the first 3D cards on PC (Voodoo, 3DLabs, etc.). The bilinear can smooth a lot your texture especially when the resolution of the screen is higher than the resolution of the texture displayed on screen after transformation.
The problem of bilinear is that it does not reduce the anti-aliasing caused by a texture with a resolution higher than the resolution of the screen. Mipmaps can solve this by precomputing smaller images.
For instance, if your image is 16x16, you will create pre-computed version of the same image but at lower resolution, like: 8x8, 4x4, 2x2, 1x1. Then you choose the best possible mipmaps for the combination of the screen resolution and the transformed texture resolution.
The next step is to use trilinear filtering which basically interpolate also between the different mipmaps to smooth as much as possible the image, even when part of the texture needs to display pixels with a too high resolution AND pixels with a too low resolution. It will use for instance the 16x16 image for the closer area and 4x4 for a farer area and interpolates between the 16x16 and the 4x4.
This is used a lot in video games (look at the ground texturing in most of the video games). However in our case, it is difficult to use that without a big performance hit. In video games, textures are preloaded and mipmaps pre-computed once. In Modul8, the mipmaps need to be generated each time a new video frame need to be displayed or a new effects is modified. We prefere using our precious CPU time for other features.
Also mipmapping is not a true solution for anti-aliasing of a full composition. The reason is that it will never smooth anything outside the texture, so all the geometry will continue to generate aliasing (borders of the polygons, situations where there is a too high density of polygons/lines). Super sampling and other technics can easily give you better results (however not always easy too implement on the wide variety of the existing video cards). Well, I talk, I talk,

...