MY EXPERIMENTS WITH AI STUFFSES

With some free AI tools available to play around with, I am trying different things with generative AI. This is my collection of such fun experiments.

Generating movie posters with randomised art styles

1. I used Bard to generate a list of 10 movies paired with random popular art styles.

2. I then asked Bard to describe what a poster of each movie would look like in the given art style.

3. Next, I asked Bard to create a prompt for the Stable Diffusion Model for each movie and art style combination.

4. Since Bard's first prompt was short, I asked it to generate a longer prompt for each combination as well.

5. Conveniently, Bard could generate everything in a tabular format that was ready to export to Google Sheets.

6. Finally, I entered all of the prompts into the Stable Diffusion XL model on clipdrop.co and added the resulting images to the same sheet to compile the results.

Here's the results

Movies x Art Syles

postersheet #,Movie,Art Style,Poster,Short Prompt (Bard),Stable Diffusion XL (clipdrop) 1,Stable Diffusion XL (clipdrop) 2,Stable Diffusion XL (clipdrop) 3,Stable Diffusion XL (clipdrop) 4,Longer Prompt (Bard),Stable Diffusion XL (clipdrop) 1,Stable Diffusion XL (clipdrop) 2,Stable Diffusion XL ...

docs.google.com

Link to the entire Bard chat if you want to see my struggles with generating the table: https://g.co/bard/share/e0f8b1c36912 (the link would expire in 6 months)

Observations

1. Table generation support on Bard is highly versatile and has the potential to revolutionize the way we gather and analyze data.

2. While it can be a bit finicky to get the prompts just right, the results are often worth the effort.

3. The difference between short and long prompts is not always clear-cut, and sometimes both can miss the mark.

4. Like all other generative AI models, Bard has a bias towards Western content.

5. When generating images of movies and characters, it is often helpful to include the actor names along with the character names.