We badly described cartoon characters to an AI. Right here’s what it drew

The usage of synthetic intelligence (AI) has made its method into nearly each business on this planet, fixing complicated issues, studying tough abilities in seconds, and usually blowing the minds of people with superior calculations we may by no means dream of.

However whereas most individuals want to use AI for the great of humanity, bettering society and fixing the large questions of right this moment, I made a decision to make use of it for a better objective – recreating a few of the most recognisable cartoon characters by means of an image-generation AI.

The way it works

Again in January 2021, the very clever individuals over at OpenAI created a program that, when fed a string of textual content, may use this to generate a picture. This might be ‘a lion on a settee’, ‘ a black gap in a field’, or another equally bizarre immediate. Nevertheless, this expertise would typically produce blurry photos, or wrestle to totally perceive the immediate that it was given. In April 2022, OpenAI launched the second iteration of this product – Dall-E 2.

Now, you’ll be able to get high-quality photos based mostly in your worded prompts, providing extremely detailed photos in a matter of seconds.

The essential factor to notice with this expertise is that, the extra data you give it, the higher the tip end result might be. Nevertheless, I lack the artistic abilities to actually paint a verbal image… and that’s fairly noticeable with my AI-generated cartoon characters. The software program is ready to carry out at its greatest when it’s given a method of artwork, and loads of description. I as an alternative opted for obscure artwork phrases and a scarcity of element – what may go improper!

BoJack Horseman

Immediate: Horse in a swimsuit as a oil portray

© OpenAI, Netflix
(Left) Dall-E 2’s picture of a horse (proper) Bojack Horseman © OpenAI/Netflix

Let’s be trustworthy, based mostly on the outline ‘Horse in a swimsuit’, Dall-E 2 nailed my description – that’s certainly a horse in a swimsuit, and my request of an oil portray model was nailed completely completely. I’d even prefer to suppose that if there was a season the place BoJack Horseman was a mid-century nobleman, my try at getting an AI to recreate him could be completely bang on.

I believe if Dall-E 2 had been given extra data on this one, we may have gotten an ideal recreation of BoJack in an oil portray!

© OpenAI
Dall-E 2’s different makes an attempt at creating horses had been simply as correct © OpenAI

There have been two different examples of horses in fits, each additionally within the model of an oil portray. I believe we will all agree that they’re simply as dapper because the AI’s different try – who knew horses may model a swimsuit so nicely?

Teenage Mutant Ninja Turtle

Immediate: Turtle sporting a black masks and holding a sword and pizza as a cartoon

© Dall-E, Paramount Studios
(Left) Dall-E’s rendition of a Teenage Mutant Ninja Turtle (proper) the unique © OpenAI/Paramount Studios

It labored…! Properly, it’s a minimum of recognisable this time. The immediate of ‘a cartoon’ did a number of the heavy lifting right here, producing a extra child-like model of the unique Teenage Mutant Ninja Turtles.

By giving it a bit extra data than I did with the horse, Dall-E was in a position to create a more in-depth try.

Whereas that is not at all a precise reproduction, it is one other spectacular try based mostly on a restricted immediate.

Donald Duck

Immediate: Duck with blue shirt and crimson bowtie as a pencil drawing

© OpenAI, Disney
(Left) Dall-E 2’s depiction of a duck (proper) Donald Duck © OpenAI/Disney

What I ought to have discovered from the 2 above makes an attempt is that the cartoon artwork model was the best way to go right here… I didn’t study. This can be a blessing and a curse as a result of, whereas that is simply the least correct try, it’s also my favorite.

These are each geese, they each have crimson bow-ties and blue shirts, they’re each even finished in several kinds of drawing, however they might not look any much less alike.

Whereas I am undecided how a lot additional data I may have fed (apart from a hat), the tip picture looks as if an apparent end result – it’s precisely what I requested for in any case.

© OpenAI
Dall-E 2’s different makes an attempt at a duck © OpenAI

Yogi bear

Immediate: Bear with a inexperienced hat and inexperienced tie as an animation

© OpenAI, Warner Bros
(Left) Dall E 2’s try at Yogi Bear (Proper) the actual Yogi Bear © OpenAI/Warner Bros

Okay, we have actually misplaced on this one. In an try and discover a Yogi Bear doppelgänger, I appear to have ended up with a St Patrick’s Day mascot.

As soon as once more, all the factors are appropriate. They’re each bears with inexperienced ties and hats, they only look nothing alike.

Not less than I used to be extra on model with the artwork model this time!

Did Dall-E 2 work?

Whereas I believe it’s secure to say this was a failed mission, I do not suppose it has something to do with the power of Dall-E 2. The software program has proved its spectacular talents throughout a number of prompts, artwork kinds and conditions, the shortage of similarity comes down to some key points.

Firstly, I went into this considering Dall-E could be on the identical wavelength as me. As somebody trying to recreate cartoon characters, the prompts are apparent to me – in fact ‘a duck with a blue shirt and crimson bowtie’ is Donald Duck! However to a program that may be very literal in nature, I’m asking for one thing fully totally different.

Artists with a significantly better creativeness than me have additionally proven that, with a extra detailed immediate and understanding of the platform, you will get some way more spectacular outcomes.

Dall-E 2 additionally has a characteristic that might have given me a way more correct end result. If you insert a picture, it should create a whole bunch of variations of it in its personal model. Looking back, a way more logical method to this activity.

Learn extra:

  • Future expertise: 22 concepts about to alter our world
  • These robotic fingertips mimic the nerve alerts of a human hand
  • Ought to we be afraid of human-like androids?

Leave a Comment

Your email address will not be published.