Stable Diffusion Sketch 1

Sketch 1

Prompt: “Pixar Style, puppy schnauzer adventurer, blue eye, tiny cute adorable, villain, unreal engine, dramatic lighting, 8k, portrait, 50mm”

Stable Diffusion is pretty incredible. It’s a multi-modal generative model, most easily used through hugging face, that you do not need to pay for, unlike DALLE-2 or Midjourney.

It is fascinating, and really highlights a question that now feels a little old and yet somehow super relevant: will there be a labour market for prompt engineers?

This image

We recently welcomed a puppy into our home - a miniature schauzer. I used that as an inspiration when learning about this model. Once I got things up and running (remakably quickly given hugging face makes things so smooth) - I started attempting to create some initial images. That led to a rabbit whole on how to elicit the best images given what I wanted. Reddit threads and youtube videos later I ended up with the prompt above.

This felt like an unguided exploration in some space the model is projecting onto. Adding words like portrait, 50mm, Pixar Style guided generation toward what I was ultimately looking for. Because this field moves so fast, espcially when things are open, there is already some neat work to package a combination of words within a prompt into a concept that isn’t covered by an english word already, for instance the style of Leica cameras.

Other generated images