Since for the classical bit representation low performance can be expected by genetic algorithms, but because of the complexity of the problem a very high performance is needed, an adapted problem representation (multimedia description language) is selected.
The multimedia description language Fib should be kept as simple as possible, with yust few alternatives, since an increases of the available alternatives will increase the number of alternatives in the application of the genetic operations and this will likely increase the amount of computation and implementation complexity.
On the other side, the multimedia description language should allow compact expressions for "normal" (occurring in the practice) multimedia objects. It should therefore provide good compression options for " normal" multimedia objects. This implies that relationships (e. g. color gradients in a surface) between parts (e. g. pixels of the surface) of a multimedia object can be represented as simply as possible.
The language should be unambiguous, reproducible and evaluable, so that an expression always evaluates to the same object. The clarity and reproducibility isn't so important for the implementation (e. g. due to different rounding errors on different architectures), if the generated multimedia object is almost always (as an example, in 
fraction of cases) very similar to the original multimedia object. But valid language objects have to be always evaluable. Otherwise restrictions on the genetic operators, by which they wher generated, have to be made or not evaluable objects could arise.
With the multimedia description language it should be possible to represent at least all raster graphics. The generation of a raster image by a Fib multimedia object in the multimedia description language should be comprehensible. That is, the multimedia description language should be designed to support the distinction of individual objects and their dependencies.
The genetic operations which change the multimedia programs should, if possible, change the program in a way, to allow a gradient descent in the hypothesis space of the multimedia programs. By running multiple operators the hypotheses, that the multimedia programs represent, should therefore be gradually improved.