Self Portraits using Stable Diffusion

Let’s start answering some previous required knowledge

What is Stable Diffusion?

Stable Diffusion is a neural network model using a deep learning for creating images from text, this task is more commonly named as text-to-image, this model comes after the news of Dalle-2 from OpenAI and Imagen from Google, both previous models are private, meaning the trained model access is private. Stable Diffusion bring something new, a trained model to public access, making accessible to public it usage, having the boom and help from people made a lot of extras to Stable Diffusion.

Learn more about Imagen : https://imagen.research.google/

Learn more about Dalle-2 : https://openai.com/dall-e-2/

How could you create Self Portraits using Stable Diffusion?

To create a Self Portrait you need first teach who you are to the model, for this come an amazing paper : DreamBooth, how so? well, Dreambooth explain a way to fine tune a model, enabling the option of adding a new subject to the model, for example your face, your dog, etc. And thank to the community there are many tutorials on how to do it. I used JoePenna repository on how to do it : https://github.com/JoePenna/Dreambooth-Stable-Diffusion

End results

It was a great experiment, there are a lot of company’s using IA today in many ways, and the part of Text-to-Image is one of my favorites, it enables a new way of Style Transfer, new kind of IA + Human Art.

Some ethical problems for the future

Right now we have advanced a lot in technology and there are some questions unresolved, for example:

Who is the owner of the model after being trained with public images of many artist’s?

If you create an image with a text prompt you created by experimenting, who is the owner of the image created?