Random result glyphs from generator

[halvomez]

Hi!
Found that the generator always produces random glyphs with the same attribute parameters passed (intensity). The font as source for generation is always the same.
Is this normal behavior?
Can the generation be constant with constant input values ​​to the generator?

[halvomez]

I also found that the trained model cannot replicate the font that is passed to the generator. For example, as part of the test, we have one font in the unmark category. I am getting the predicted attributes through the model and passing them to the generator. That is, the difference between the intensity is 0. At the output, I will only get a similar font to the original one, and each time it will be random. Thus, if I want to change only one parameter in the original font through change one attribute, for example wide, then I still will not get wider source glyphs, I will just get different ones

For "train + valid" category result is better since the attributes are derived from manual markup

If I want to get as similar as possible to the original font, but only with minor changes, what can I do?