A Variety of DALL-E Dollies

Today started out totally normal for me.  I got up, let the dogs out, grabbed a cup of coffee, and sat down to read the news.  But then an article in the Washington Post completely changed the course of my entire day.  And forgive me in advance, but it's likely to steal all of the free time you thought you had, too.

The article I read was about DALL-E, an artificial intelligence, text-to-image generator that has just become widely available to the public.  A text-to-image generator is software that can use regular text to create an image.  So if I input a phrase like "pencil drawing of a dog," DALL-E will give me several computer-generated images that look like pencil drawings of dogs.  And judging by the Washington Post sample images, I could see that DALL-E is very good at what it does.  The article was focused mainly on how this type of AI can be dangerous--and I can see that angle.  For example, people wanting to make a point could generate fake photographic evidence to claim something happened when it actually didn't.

But of course I wasn't interested in DALL-E because I want to create conspiracies or topple governments, I just wanted to see how good this thing is at rendering photographs of dolls.  And it's darn good...at least some of the time:

DALL-E redhead doll in a blue polka dot dress.

That beautiful redhead is one of my favorite images from the six hours I spent on DALL-E today.  She's lovely, and aside from her one wonky eye, I wish she was available in real life for me to buy!  But, as you'll see, my results were not always that good.

In order to start using DALL-E, all you have to do is go to their website and set up an account.  This involves giving them your email address and phone number, choosing a password...and that's about it.  With free membership, you get 50 credits to start with, and then 15 additional credits every month after that.  And of course at any time you can buy 115 credits for $15, which I think is a pretty good deal.  I blasted through about 80 credits today alone.

Each credit gives you the chance to type a phrase that generates four images.  The images are all different interpretations of what you typed.  

For some reason, probably because of recent cat doll reviews, the first idea I got was to try and generate a doll with a Sphynx cat head.  So I typed, "an articulated fashion doll that looks like a hairless cat."  And this is what I got:

Freaky!

And a more sultry version:

I was pretty impressed by these, especially since it was my first attempt!  The images can have some strange things going on, though, as evidenced by this next kitty and whatever in the world is happening near her cleavage:

And then sometimes DALL-E misses the point completely:

Definitely not a fashion doll.

I explored this avenue a little further by trying to make a cat doll with a fuzzier head--like that of a Siamese.  I tried the phrase "portrait of a Siamese cat fashion doll with blue eyes and a long tail, wearing a tight black dress and licking her paw." 

Something got lost in translation:

I think the part where I said she should be licking her paw threw everything into crazy land.  

This next picture looks like the cat is licking a kitten head.  Or is that supposed to be her paw?  I'm so confused.

Where's the rest of that kitten's body?

The pictures in this group just kept getting worse and worse.  

Like what in the name of all that's good is happening here?

Whelp, there's the rest of the kitten's body.

And then this cat, rather than wearing a black dress, simply has another cat growing out of its back:

A Siamese twin, I guess?

If things go badly like this, you can modify the phrase you wrote and try to fine-tune it, thank goodness.  I tweaked mine to read "portrait of a Siamese cat fashion doll with a long curvy tail, wearing a tight black dress, and licking her hand."  

Apparently the substitution of "hand" for "paw" made a difference...but oh, DALL-E, sweetie.  That's not how tails work:

The other images in this group were better, though, and I was finally getting what I asked for:

Too bad the top of her head is cut off.

This one is practically purr-fect...as long as you ignore the melting left eye:

Nailed it!

My luck was not as good when I tried to invent dolls with other types of animal head.  I wanted a Barbie doll with a dragon head, a long black dress, and bat wings, and I got this:

Something about asking for a dragon head on a doll broke the program and made all of the faces into indistinct blobs:

The wings are cool though.

This one's face is being swallowed by her neck, and it looks like she has metallic hair!

Terminator Dragon Barbie.

And this one's face looks like something from Star Wars:

But that dress is fab.

I also tried making a rabbit-headed doll by typing, "an art doll with a human body and a cute rabbit head with large blue eyes, sitting on a log."

I didn't quite get that:

"Bored young lad draped over log."

This one probably comes the closest to what I asked for, with some actual rabbit ears growing out of the head:

But this one is my favorite:

With legs going any old way.

I pushed my luck by attempting a doll with a corn muffin for a head (Patreon members will know why...), and poor DALL-E could not grasp the concept.

I typed "a photograph of a doll with a human body and a corn muffin head, with large hazel eyes, sitting on a plate."  

But what I got is "melty-faced girl hovering on top of a bagel:"

Okay it's not even a bagel, but what is it??

The next doll has a strange face with double eyebrows on one side.  She also has a bizarre hairstyle that maybe we should call a "corn muffin top ponytail?"

But that's not a corn muffin on her plate.

This doll is at least sitting on something that could be an actual corn muffin, which is real progress, and she's pretty cute:

I love her tutu.

But this one steals the cake (or the muffin):

Grouchy Agnes, on the candlestick, with the muffin.

After a while I gave up on trying to put strange heads onto dolls and tried my hand at creating flocked animal characters--you know, like the Honey Bee Acres unicorns.  

I typed, "a flocked unicorn animal toy with jointed arms and legs and rainbow mane and tail standing inside a play store diorama."

DALL-E dismissed the idea of a play store diorama and just gave me a cute fuzzy unicorn:

Fine by me.

These images are all delightful, but they don't remind me very much of the Honey Bee Acres critters.

I especially like this one:

And this one is very dramatic, with that flowing mane and raised hoof:

I tried adding that the unicorn should have a human-like body, and this is what I got:

A rolling pin arm.

Something about the shop concept got through to DALL-E at this point, too, for some reason, and the images suddenly all had shelving in the background:

The shopping warrior!

I tried to describe a few photos where a real animal was interacting with a doll, but DALL-E was very reluctant to show me the dolls' faces:

...with one exception:

I'm not sure the doll is the one who should be freaked out.

I also tried experimenting with dolls riding on horses, but the photos all cropped out various parts of the doll or the horse, which was awkward:

Interesting leg position there.

No head for this girl:

And she's grumpy about it.

And no mouth for this one:

Mmmph!

I suggested that the riders should be smiling, to avoid them having no mouth whatsoever, but the smiles got unnervingly sinister:

I never said anything about demon clowns.

So I deleted the smiling part...and got a lot of judgmental glares as a result:

The horseback riders reminded me of my Nendoroid Link review, so I tried to describe that figure by typing, "a Legend of Zelda Link action figure riding on his horse Epona with a bow and arrow in his hands."  

But I don't think DALL-E knows what The Legend of Zelda is...not completely, anyway:

This is Captain America meets Link:

Riding a malnourished horse.

Some of the images made me think that a tiny bit of data about Link was being used.  Not only is there green in all of the figures' outfits, but the blurry head on this one has some vaguely Link-like attributes:

And somebody twisted that poor horse's nose all around!

I like how even though I only said that the figure should be "holding a bow and arrow" many of the images had cool action poses like this one:

Never mind that he isn't holding the arrow.

Once I was done with my animal experiments, I decided to go shopping in the DALL-E doll store, to see what kinds of normal dolls are available, by typing "a toy aisle with a shelf full of fashion dolls in bright boxes:"

No normal dolls here.

These images are wonderful from a distance, and almost look legitimate, but they're terrifying when you start to look up close!

Is that a little girl standing on the left?  Yikes.

These princess dolls look like they have beautiful dresses, but the mish-mash of facial features is disturbing!

And these dolls don't have enough box space to be hiding complete bodies...but they're not exactly styling heads, either:

What are they?!

Anyway, I typed some text that I thought would give me the kind of doll that I typically buy, namely "a jointed play doll with curly red hair and freckles, wearing a blue sailor dress and a blue dress coat."

Not bad, DALL-E, not bad!

That's a very toothy grin, though.

This one even has a blue dress coat, sort-of:

I tried some different outfit concepts and got several truly lovely results!  

One thing I noticed is that if I specified a color for the dress and a separate color for the eyes, DALL-E would often pick one color and use that for both the dress and the eyes.  This is nice, I guess, because it means the doll's eyes always match her outfit.

One detail I had to delete after a while was the freckles, because DALL-E does not seem to completely understand how freckles work:

Cat scratch face ≠ freckles.

But feast your eyes on all of these beautiful (slightly strange) creations!

Christina Ricci's Wednesday Addams, for sure.

She looks like Evangeline Ghastly!

There are several of those photos that, if the dolls were real, I'd be interested in buying them.

I also tried to create a few scenes that mimic my end-of-review outdoor pictures.  In this first series, I tried to give the dolls a living animal companion (a squirrel), but DALL-E decided it should be a toy squirrel instead:

Or a little warthog.

This looks like a painting!

That's a strangely adorable squirrel.

I'm not sure this is a squirrel, though:

It's a baby jackalope.

I tried again with a butterfly instead of a squirrel, but all of the dolls seemed scared!

Or suspicious:

Or simply depressed:

I tried for some action scenes outside, too, by saying that the dolls should be running in the woods, and this started out really well.  

This girl is like a wood nymph on the hunt!

But then things deteriorated...

Until we got to this bizarre creature with her unnatural posture, no mouth, and maybe an eye on her forehead?:

I set up some other very specific scenarios, too, like "a dramatic, realistic fashion doll with red hair and long legs sitting in a modern chair with a single light hanging overhead."  

Close your eyes and picture that scene for a moment.  It's very elegant, right?

Now look at this:

I never said to take away her face!!

This doll is judging me:

And this one is terrified of that enormous light!

And somehow she got short legs and robot feet.

This one is just being naughty:

Peek-a-boo!

Incidentally, DALL-E won't allow anything that's intentionally explicit or gross.  Like I wanted to recreate one of my articulation photos, where the dolls aren't dressed, and that was not allowed.  Nor was anything to do with dog poop.  Just saying.

I shifted gears and tried for something more cheerful, like "a doll with green hair and elf ears and big fairy wings, sitting outside in a field full of colorful flowers."

Cute right?  This one even has a unicorn horn, although I'm not sure how that happened:

Or maybe that's an elf ear.  Let's not think too hard about it.

And this one got bug-like antennae:

And buggy eyes!

This one is the most realistic:

Does she have a bindhi?

But what in the world happened to this poor girls toes!

The same thing that happened to her left eye.

I tried for a more mundane set-up, typing "a Barbie doll with red hair shopping at the grocery store for oranges."

DALL-E is aware of scale at some level, which I find fascinating.  This 1:6 girl is shopping for full-sized oranges:

In a groovy outfit!

This next doll's face is intriguing, and I love her outfit, but there's something messed-up with her hand:

I didn't specify anything about the outfits for these images, but I like how all of the dolls are wearing different, fashionable things!

Some of my ideas went horribly wrong.  I thought it would be fun to include a baby doll, so I did my best to describe a crying baby--picturing this kind of thing:

A Berenguer baby.

But with the seemingly innocent description of "crying baby doll with an open mouth and big eyes, wearing a fuzzy blue romper," I got this nightmare:

And the others in the series were equally horrific:

I most certainly did not ask for a melty-faced baby with his mouth caving in and hands where his feet should be!

I'd cry, too, if my legs were bent like that.

To erase that nightmare, I tried for a happy baby doll sitting in their crib, holding a toy: "a smiling baby doll sitting in a crib and holding a toy."  

Instead I got a realistic-looking baby holding a doll!

The interpretation of real babies is uncanny.  Some of them look like they must be real people:

While with others, it's clear that there's something amiss:

I was curious to see how aware DALL-E is of certain well-known doll brands and styles, so I typed in things like "a photograph of an American Girl doll."

DALL-E does not know what an American Girl doll is:

That looks more like a Kurhn doll.

I got a really nice assortment of dolls with this text, though.  They all look real, and they look like dolls that I would buy.

She kind-of has American Girl stars on her sweater...

This cutie has an American flag draped around her:

DALL-E also has no idea what a styling head is.  More like a stylin' head:

The software does seem to know what ball-jointed dolls are, though.  I typed, "portrait of a ball-jointed BJD doll with silver hair and grey eyes, wearing a lace shirt," and this is what I got:

Pretty convincing, right?

Although DALL-E really struggles with getting those eyes to match:

Nice try.

I was also able to successfully generate a rag doll...although she never really wanted to be centered in the photo!

This one is my favorite:

I tried describing a "fashion doll that is made out of LEGO building bricks," and that worked very well!

Except for maybe the hair.

Cool dress!

This one is even striking a pose:

And sticking out her tongue.

Next I tested to see how well DALL-E understands the world of Harry Potter.  I asked for a "Hermione Granger doll, posed with one hand holding a wand up near her head."  

There are some issues with the color palette, but the basic concept is there:

She's cute!

I love this one's focused expression:

 I modified my text to include, "wearing the Gryffindor school uniform," and things improved slightly:

The colors are right, but their distribution is all wrong.

And this Hermione managed to make her own nose disappear:

Nasus evanesco!

DALL-E seems to have an excellent idea about what a Blythe doll is:

Epic glare.

I don't know what's sticking out of the doll's dress, above, but this next doll appears to have mushrooms on her dress, which I did not ask for!

Some of the photos I got during this experiment reminded me a lot of photos I've taken myself.  Sometimes I was trying for that effect, with the grey background or the dolls leaning against trees, but not always.  I mean, DALL-E is almost certainly incorporating this blog's photos into its database, which kinda blows my mind.

Anyway, despite all of my photos of My Twinn dolls, DALL-E has no idea whatsoever what a My Twinn is:

Nope.

These are the results from me typing "a My Twinn doll with dark hair and grey eyes, wearing overalls, with one hand raised above her head in a wave."

The resulting images are cute, but they're nothing close to a My Twinn.

Nice try, DALL-E.

DALL-E is a bit of a Star Wars expert, though:

This is "Star Wars Princess Leia with her hair in two buns, posed with one hand raised holding a blaster, with the Millennium Falcon in the background."

So DALL-E figured out the iconic white outfit all by itself.

That's someone's hand in the background, though.  Holding a remote?

The blasters are a little sus in these next two photos, but Leia herself is great:

Nice facial features!

At first, Monster High appeared to be a totally foreign concept to DALL-E.  This was surprising to me, considering how many Monster High photos there must be on the internet at this point. 

I tried for a generic Monster High doll, dictating only the color of her skin (and therefore her hair):

I typed "a Monster High doll with green skin and pink hair."

That first shot is not a terrible attempt, honestly, but there was not much consistency:

And I couldn't really hone in on any specific character:

I tried to approximate Ghoulia by dictating her skin color and red glasses:

And some results were better than others:

Then I tried asking specifically for Clawdeen Wolf, and this gave me a fascinating mix of results:

Maybe DALL-E knows something about Monster High after all, because these are too good--especially given the trouble I had making photos of dolls with other kinds of animal heads.

And they got her color scheme right, too, which I did not convey.  I just said "Monster High Clawdeen Wolf standing against a grey background."

They got some of Clawdeen's sass, too!

Geez, look at those calf muscles!

Disney-related text was very rewarding with DALL-E, with the exception of asking for "a Mirabel doll from the movie Encanto."  

That was a swing and a miss:

Why is there a foot sticking up at the bottom of the image??

She's worried about that foot, too.

But there's definitely a Disney vibe to these images, which gave me hope!

So I tried asking for "a portrait of a Belle doll from Beauty and the Beast, wearing a yellow ball gown."

Ta-da!  She's pretty great!  Just don't look too hard at her mouth:

This next one's dress is even better, but I think it's a Once Upon a Zombie Belle:

Braaaains. Guessst.

Elsa from Frozen is easily recognizable, too:

Although I specified her long blonde hair, the blue dress, and the ice castle in the background.

This next one looks like a Disney Store Rapunzel doll to me!

Cute, recognizable face.

Merida is not as easily recognizable:

What is her right arm doing??

Although the facial expression is spot-on in this image:

And DALL-E seemed to understand that "Ariel from The Little Mermaid" should have a tail, even without me saying so:

But I had to specifically describe the backdrop of a rock with a crashing wave.

I mentioned nothing about Ariel looked completed pissed off.

Nor the squid stuck to her head.

I made a few alterations in honor of our new Ariel, too!

Her face is blurry in all of these, but the drama of the scenery and posing is excellent!

And you knew where all of this was leading, right?

It was a bit of a challenge to make DALL-E generate Cinderella in her rag dress, but I got a few examples.

But when I asked DALL-E to please show me "a Disney Cinderella doll with blonde hair, blue eyes, and a blue sparking dress with butterflies..."

Things really took off!

Brown eyes are fine, too. No biggie.

I wish these dolls were real!!

Especially this one.

And this one!

...but maybe not this one:

But this version has such an endearing expression:

And while this one's eyes are a little smudged, she's still lovely:

Honestly, I don't know how I'm going to avoid generating a hundred new fake Cinderella dolls every day.  I won't admit how many I made just today.

Hint: I'm number twenty three.

Here's another one of my favorites:

Ignore the mismatched eyes.

That's probably enough.  You get the idea.

Okay, to finish things up for today, I know probably a few of you have wondered over the years what I look like.  I don't really splash my face all over the internet.  So I figured this was a good time to finally show you.

This is me:

What are my hands doing...and how many do I have?

No, just kidding.  This is me:

Minus the bangs.  I don't have bangs.

But this is me before coffee, with my mini me:

And this is me getting mad when a doll won't look at the camera.

This is me most days when I can't decide what to review next:

And this is me after spending my whole week with only the company of dolls and dogs:

You knew I had two rows of teeth, right?

But this one sums me up the best, and I'd like to adopt it as my official portrait:

Bottom line?  That was a very strange day.  Thank you, DALL-E, for the laughs.

And all of you, what are you still doing here?  Get yourself over to the DALL-E site for a free account and start having some fun of your own!  I'll see you in three weeks when I've kicked this addiction.