#4 Training a diffusion model with only handwritten fonts

Experiment #4 aimed to generate As in the style of Handwritten fonts. I created a dataset with about 1,300 As in fonts of the Handwritten category. The effect of input and output can be seen clearly: In the Handwriting category, there are more diverse fonts, and so the output also varies between sketchy lines and broad brush strokes. But here, too, the model has "learned" in a few tens of thousands of training steps what a handwritten A might look like.