In general text is not generated artificially because this leads to non-realistic datasets. In the case of LDA it would be very easy to generate data using LDA itself since it's a generative model. However this would make it a much easier job for LDA to estimate the parameters than with some real corpus.
So as far as I know most experiments about topic modeling are made with some real corpora, for example the UN corpus, the State of the Union corpus, the Europarl corpus, etc. The advantage with topic modeling is that there's no need for annotation so any large collection of text can be used.
Does anyone have any methods for generating data for an LDA model? Where I can control topic number, document number etc?
Note that the number of topics $k$ is a parameter in LDA, so whatever the data LDA searches for exactly $k$ topics. The number of documents is fairly easy to control if you use any large collection of documents. The main difficulty with topic modeling is how to evaluate the resulting model.