4.
Extracting, Prompting & Coding

> In a first step, I had descriptions generated from all 700+ images. The descriptions are extremely accurate and are probably just as detailed as if I had made them. The image descriptions were created with the new ChatGPT4. 
> My aim was to automate the processes as much as possible. I therefore made about 15 different attempts to completely automate image generation. 
> Also I tried many different A.I. tools for text-to-image generation such as Stable Diffusion, Dall-e etc. 

> Completing the automation was technically quite difficult and took a lot of time, but I saved a lot of time by automating this process. 
> I created the automation with the help of Copilot and ChatGPT in the form of a Python3 script. 
> The script is relatively simple, the prompts are loaded from a text file, then sent to Dall-E using the OpenAI API (programming interface) and an image is returned and saved locally.