The super cool artificial text to image generator that took over the internet has just been open to the general public on September 28. I am here to start using it for the first time and blog about my general first impression of it.
First we go to the post by DALL-E and see how we can sign up.
DALL·E Now Available Without Waitlist (openai.com)
Signing up was pretty straight forward. Click on sign up and create a quick account.
Then I get to a page where I it tells me to just enter a very detailed sentence to get the best results. Below are the images and the text I used. Let see close to meeting the hype.
Generating Images
6 dogs playing poker and wearing steak hats.
A banana bed in a fantasy world.
A picture of yourself with a subtitle to give a world a message you want us to know.
A robot painting a artwork of a man eating sushi.
2 androids creating a new version of themselves in their secret lair.
A corvette F-35 fighter jet flying out of a fantasy dragon in mario world.
super saiyan mario
My initial impression
Based on the articles I have read about DALL-E earlier in the year, the results are disappointing. It looks like the results are definitely random and you may have to use the same text over a few dozen times before you get an image of what you want. This is probably why DALL-E opted for a pay per click subscription.
I would say this would be great tool to teach kids about how machine can take your input and try its very best to understand what you want. Also that depending on what you are trying to make it paint is highly dependent on the image data sets it was given.
Definitely a neat tool, but I would not rely on it to help you create consistent artwork over and over again. It would work if you were looking for a general concept and you don’t mind spending a day with trial and error. Which may still be faster than trying to draw these images yourself.