Trouble training a style

[carldaws]

First of all, thank you for this implementation. I'm hoping you can help.

I have a specific set of illustrations and I'd like to generate more in that style. I've run the training process twice, once using a description of the style (linear illustration style) in the caption and once using a rare word (sws style) in the caption. The StyleDrop paper seems to indicate that both methods would work, however neither produced very accurate results. The description of the style worked best, but it seemed to only generate 'linear illustrations', not ones that looked particularly like the images I supplied. Using the rare word ('sws style') approach seemed to do nearly nothing - it seems almost like the images are being ignored and only the caption is being considered. Am I doing something wrong?