This is a model trained using my own hand-drawn sketches. The aim is to create a reliable means of generating loose sketch images that focus on capturing the key details of a scene in a simple and striking way.

The LoRA was trained for 10 epochs using Kohya.

For best results I recommend using the following positive prompt:

<lora:SKETCHY_JBX:0.8> SKETCHY_JBX a sketch {whatever you want to see here}

At strength 1 expect to see a very loose style that might be a little overpowering. Best results seem to sit between 0.6 and 0.8 - at lower values the sketchy nature will remain but the model will start to colourise some images in a manner that looks a little like watercolour.