Table of content #

What is this all about? #

We were curious about how much variance the AI has. So, what would be the results if we were to request 100 images with the same prompt? - I won't review the results and rather just present the results to you.

These prompts are a result of a quick brain storming. If you have suggestions, please let me know. I might create more posts like this in the future. The goal was to have a wide range to motives, styles, and so.

These images are unedited. Generated - downloaded - created a montage; that is it. These images are free for personal or commercial use and do not require any form of mentioning. Dall-e gives ownership of the images to me, and I give you permission to do with it, whatever you want.

The resolution of the originals is 1024x1024 and I might provide a download link at some point. If you want a single image, feel free to reach out.

With testing, the total costs were around 20 EUR. I'd say that it is acceptable.

You can find a technical write-up at the end of the post. But as a disclaimer: not best-practice. Feedback is still appreciated.

Gallery

So, enjoy!

1 - Cats #

photo of a kitten on a carpet in the living room, digital art

cats

2 - Robot #

small robot wandering around in an post-apocalyptic world, digital art

robot

3 - Donut #

minimalist logo of a donut shop

donut

4 - Dackel #

dackel in a suit in a library, digital art

dackel

5 - Poster #

movie poster for an action movie from the 80s, digital art

poster

6 - Citylife #

a black and white photo of the life in new york

citylife

7 - Dolphin #

sticker illustration of a cute dolphin

dolphin

8 - Light #

area view of a city with street lights at night, digital art

light

9 - Monster #

detailed sketch of an evil monster, digital art

monster

10 - Cyberpunk #

realistic photo of a colorful cyberpunk city in the rain at night, digital art

cyberpunk

Tech write-up #

Side note: To be clear, this is not best-practice. It got its job done, and that is all I needed. Still, feel free to reach out, happy to learn!

First, openai Dall-E API offers to generate the following sizes, with 3 different prices:

Resolution  Price
1024×1024   $0.020 / image
512×512     $0.018 / image
256×256     $0.016 / image

I've generated the largest resolution.

Limitations #

So, I've decided to use the API via curl, the first limit I encountered is the '10 images per request'.

{
  "error": {
    "code": null,
    "message": "20 is greater than the maximum of 10 - 'n'",
    "param": null,
    "type": "invalid_request_error"
  }
}

The next one would be the rate limit of 50 images per 5 minutes.

{
  "error": {
    "code": null,
    "message": "Rate limit reached for images per minute. Limit: 50/5min. Current: 60/5min. Please visit https://help.openai.com/en/articles/68839691 to learn how to increase your rate limit.",
    "param": null,
    "type": "requests"
  }
}

In the end, the download of the generated images was limited too. After every category I had to switch to another VPN server location to bypass the limit.

Script to download them all! #

I did a small break after every category to check the result of the script, and whether all images were generated and downloaded.

I'll add some comments later, but in short:
generate images and put curl response to file
get URL from output file and remove the quotation marks "
download images via curl
wait one minute to avoid rate limit
#!/bin/bash

# For for-loop for the whole script due to the limitations
# Curl request to generate the images via API, and the save the output via -o flag to a file

for i in {1..10};
do
  echo $i

  curl -o output.txt https://api.openai.com/v1/images/generations \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer sk-sdsdskdsdsdsdeefefe" \
    -d '{
      "prompt": "small robot wandering around in an post-apocalyptic world, digital art",
      "n":10,
      "size":"1024x1024"
    }'

# Gets the URLs of the generate images, removes quotation marks, and saves it to a new file (one URL per line)
  cat output.txt | jq '.data[].url' | sed 's/"//g' > output_url.txt

# Finally, download images with curl to the current directory. I was told that this is not bet practice, but it worked.
  cat output_url.txt  | while read f; do curl "${f}" -O; done;

# wait 60 seconds before we start it all over again
  sleep 60
done

Things to improve: start/stop, logs, error and information notification, speed

Rename everything #

In the next step, I had to rename all the files. The file names were cryptic and difficult to work with.

#!/bin/bash
a=1
n=cats
for i in ./1_cats/*; do
  new=$(printf "./1_cats/"$n"_%04d.jpg" "$a")
  mv -i -- "$i" "$new"
  let a=a+1
done

The name scheme would look like: cats_0001.png

Create montage with imagemagick #

In the last step, I used imagemagick to create a montage with the following command.

montage -geometry 200x200+2+2 -tile 4x -set label '%f' *.jpg montag.jpg

Explanation:
montage # imagemagick function to create montages
-geometry 200x200+2+2 # size per image + min size of the padding between the images
-tile 4x # setting for the layout, 4 columns, unlimited rows. 3x4 would be a limit of 3 columns and 4 rows
-set label '%f' # adds the filename of the image on the montage
*.jpg # use ALL .jpg file within this directory for the montage
montag.jpg # name and format of the final montage