AI Image Generation

This is a presentation I am sharing with my students on AI Image Generation. I will talk through the slides on this page. Here is a link to the presentation AI Image Generation.

This is my title slide. The Art Work was created by Stable Diffusion in DreamStudio using the text prompt: “A poster on AI Image Generation by Banksy”.

This picture was submitted to the Colorado State Fair and won first place in the Digital Art category. When the artist, Jason Allen, went up to collect his prize, he said that this image was generated by AI image generator Midjourney.

Is this fair?

This is where I want to have a class discussion.

Greg Rutkowski is angry. He is a talented artist who has made a name for himself creating art for video games and Dungeons and Dragons. This piece of art in the slide is from the video game Horizon Forbidden West. His name has been used 93,000 times in text prompts for different AI Image Generators. This is far more than any other artist. His work and style are being copied by different AI models. How is this possible? The AI models have been trained on huge datasets and they have been trained on Greg Rutkowski’s artwork.

This is where I want to discuss how to protect artists work.

This is a relevant quote from Greg Rutkowski.

In order for artists to find out if their artwork has been used to train popular AI art models, the website Have I Been Trained? was set up. One artist, Lapine used this website to see if she had been trained.

When checking to see if her artwork was in the dataset, Lapine discovered that her private medical images were a part of the dataset. She had expressly asked them to be kept private.

This is another opportunity to discuss the ethics of these datasets. They are created by bots scrapping the Web for any available images. Who is responsible in this situation?

This is the interface for DreamStudio. DreamStudio is a web page that allows you to access Stable Diffusion. Stable Diffusion is an AI Art generating model released in August 2022. Once you sign up, you will have 200 free credits to experiment with generating images.

This is an example of a text prompt used to generate these three images. Adding an exclamation mark, makes that part of the prompt more important for the AI model. A few different artists have been named to generate these images.

At this point I will ask the students to create some AI Art of their own. I will ask them to use DreamStudio. I will also ask them to research some text prompts using Lexica and Visualise. When they have created some image, I will ask them to upload them to a Padlet with the text prompts that they used.

If you get interested in AI Art and run out of credits on DreamStudio, you can download Stable Diffusion on your computer if your computer meets the requirements. Alternatively, you can use a Google Colab. A Google Colab is a notebook that allows you to run Python code in your web browser. Google allows you to connect to a GPU to run your text prompts through Stable Diffusion. Here is a Google Colab that runs Stable Diffusion.

Google Colabs also make it possible to incorporate DreamBooth into your creations. Google’s DreamBooth allows you to train a model using images uploaded to Google Drive. You can then use this model with Stable Diffusion. In this example, I created my cat Kiwi and then used Stable Diffusion to put him in many different locations.

This is another example of using a Google Colab. In this case I used the Stable Diffusion to animate based on a text prompt. I found this tutorial very helpful. The music is from my nephew who is very talented.

Here are some citations for my slides. I hope you found this useful and you can use some ideas with your students. If you create any AI Art, I would love to see it.