MY EXPERIMENTS WITH AI STUFFSES

This is a series of posts to share my experiments with generative AI. It's a compilation space for my random, fun experiments with generative AI.

Asking Bard to describe an image to then generate another image and checking the understanding of Bard again in order to generate another image and so on...

Started off with this source image, which I asked Google Bard to describe for me.

Using the descriptions, I asked Bard to create a prompt in order to recreate the image. This is what it came up with:

✒️Bard generated prompt: A city street with tall buildings and cars driving down the street. The cars are parked on the side of the street and people are walking on the sidewalk. The sun is shining and the sky is blue. The buildings are a mix of old and new, with some skyscrapers and some smaller buildings. The street is lined with trees and there are a few benches where people can sit. The overall impression is of a busy, bustling city street.

1st Degree Result

Using this result, I asked Bard to describe the picture and generate another prompt to recreate this picture again. This was the next prompt:

✒️Bard generated prompt: Generate an image of a city street with tall buildings and cars driving down the street. The cars are parked on the side of the street and people are walking on the sidewalk. The sun is shining and the sky is blue. The buildings are a mix of old and new, with some skyscrapers and some smaller buildings. The street is lined with trees and there are a few benches where people can sit. The overall impression is of a busy, bustling city street.

2nd Degree Result

Using this result, I asked Bard to describe the picture again and generate another prompt to recreate this picture. This was the next prompt:

✒️Bard generated prompt: Generate an image of a group of people walking down a city street. The people should be all different ages and ethnicities, and they should be all dressed in casual clothes. The street should be lined with trees and there should be a few benches where people can sit. The image should be realistic and detailed, and it should capture the hustle and bustle of a city street.

3rd Degree Result

I could've gone on, but I stopped here.

Observations

1. Bard was able to provide a decent description of the image and create a prompt that was relatively accurate.

2. Even though the first two prompts were almost identical, the generated images were significantly different in quality. This demonstrates how much variation can exist in the output of a generative AI model, even when the prompts are very similar.

3. It was interesting to see how the image was interpreted slightly differently each time, resulting in images that were quite different from the original but still retained some elements of the initial image, such as the buildings, street, and weather.

4. If I knew how to script, it would have been a fun exercise to keep looping this process to see where a source image could go.

5. The impressive results I achieved with such minimal effort in a very short time show how powerful this technology is. I could have continued to tweak the prompts to try and get even better outputs, but I am already excited to see where this technology goes.