Artificial Intelligence and Quilts — Revisited

On July 17, I published a blog concerning Artificial Intelligence and Quilting.  I received quite a bit of response about that blog – which I hoped would happen.  I also promised to return to the AI program I used to create the images in a month.  Since AI continues to scour the internet long after a request is given, constantly searching for ways to redefine the images and information – hone and make it more accurate – I thought it would be interesting to revisit DALL.E, the AI image program I used in the first blog, and see what it would do with the exact, same image requests.

The first image I asked for in the July blog was a quilt made from 150 half-square triangles.  This is what a half-square triangle quilt can look like. It’s a pretty basic design.  I did this on purpose because I wasn’t sure what DALL.E would return.  The July 17 DALL.E rendition of a half-square triangle quilt was this:

A month later, DALL.E returned  two graphics of a HST quilt.

In the first quilt, you can kind of see where the AI program was still pulling information together.  While not quite as “off” as the July quilt, it still appears the program is throwing halved squares and triangles together in an attempt to come up with a quilt made from 150 HSTs.  It’s kind of right, but not what a quilter would expect.

The second HST quilt is a bit more cohesive.  This quilt has the half-square triangles (although not the requested 150) and has used those to create flying geese – something which kind of surprised me.  It also employs some complex piecing.  Don’t believe me?  Take a good look at the center medallion. It has a square-in-a-square, as well as some inset seams. 

And have you noticed in both of these quilts, DALL.E provided quilting stitches – something it didn’t in all of the July blog pictures. 

The second quilt I requested in July was a New York Beauty, as shown below.

DALL.E returned this quilt, which is hilarious.  Boats and taxis on the same street.  Nothing at all resembles a New York Beauty.

However…this time DALL.E did its homework. The first quilt it returned in was this:

At least the taxis were on the street.  I’d have to mark it down for workmanship.  The corners are chopped off squares.

The second New York Beauty returned was this:

It’s easy to scratch your head over this one and wonder why AI would think this quilt would ever resemble a New York Beauty.  However, I do see a resemblance of spikes coming from a center circle.  I actually think this is a lot closer to a New York Beauty than the other two.  With this quilt, you can tell AI is scanning the information available, working through it, and coming up with something.  I think this is the quilt I would be interested in coming back to in a year to see what DALL.E would return then.

I also asked DALL.E to create a Marie Webster quilt.  Here are a couple of examples of her quilts.

I love Marie Webster quilts!  The applique is just brilliant and there’s plenty of room for wonderful quilting to up the texture game.  When I asked DALL.E to render a Marie Webster quilt in July, this is what it gave me.

When I asked it again for two Marie Webster quilts, this is what it returned.

While there is some applique on the above two quilts, it’s really nothing like the delicate, floral designs which Marie Webster produced.  And she was famous for her use of soft pastels.  There are no pastels in the newly rendered AI quilts.  Which led me to ask the question:  Why, with the other quilts becoming so close to what they really need to look like, did Marie Websters end up so horribly wrong?  I think I have that answer, just hang on to the end of my blog.

Finally, I asked DALL.E to produce two quilts made by Sherriquiltsalot.com.  These are a few of my quilts. 

I really don’t have a distinct “style” like some quilters do – the closest I come is I tend to have quite a bit of orange and yellow in my quilts and I do use a lot of applique (it’s my favorite technique). 

This is the quilt DALL.E returned in July.

To be sure, these are definitely my colors, and there is a floater in one of the quilts.  However, it’s hopelessly chaotic and it doesn’t have any applique.  This time DALL.E was a bit better.

This quilt does look like something I would make.  The pinks are one of my colors of choice and it’s not as chaotic as the first quilt.  Quite possibly this would be my choice for a quick crib quilt for a baby shower. 

It’s the second quilt I’m most dubious about.  To be sure, there are some tiny squares, and I do enjoy those, but there is very little piecing.  For a woman who once sewed 45 pieces into a 5 ½-inch unfinished quilt square, the fact that there is little piecing is a bit insulting. 

Finally, the last quilt I asked DALL.E to create in the July blog was an applique quilt with a black and white cat with tulips and daisies.  I’d like to make a small quilt with a cat which resembles Felix.  I was quite pleased with the two that AI came up with:

When I asked DALL.E this time for the same type of quilt, this is what it gave me:

I like these equally as much as the first two, although the new renderings are a bit more stylized than the first quilts.  I could be happy making any of these cat quilts.

In my opinion the cat quilts are the best AI quilts.  And this got me to thinking.  In the other quilts, I gave some really specific parameters – names of quilts and a designer.  With the cat quilts, I gave general directives:  black and white cats, tulips, and daisies.  The cats had to be black and white, but the tulips and daisies could be any color.  Keeping general directives in mind, I returned to DALL.E and asked it to create a quilt with four-patches and birds

While I don’t particularly think this quilt is attractive in this color way, the four-patches are there and so are the birds.

In this second quilt, there are no four-patches, but the birds are really cute.

Maybe DALL.E doesn’t like creating pieced quilts?  It certainly seems to have no problems creating applique quilts. 

All of this experimentation has been a lot of fun.  But it still makes Artificial Intelligence no less of a threat if used incorrectly.  My fear, like a lot of creators, is that AI could hijack the work of a quilt designer.  There is the possibility AI could design quilts similar to those of our wonderful quilt designers and cut into the income of these folks.  Which also led me to wonder if AI could design fabric the same way?  Could AI possibly come up with similar designs to Buttermilk Basin, Henry Glass (both of which are fabric house with lots of designers), or Kaffe Fassett? 

So I tried.  I asked DALL.E for  two designs of each.

This is it’s rendering of Buttermilk Basin designs.

This is a swatch of the actual product.

Here is DALL.E’s idea of Henry Glass Fabrics

Here is a swatch of actual Henry Glass Fabric.

And finally, here’s AI’s idea of Kaffe Fassett’s fabric

Here are some swatches of real Kaffe Fassett fabric – DALL.E wasn’t too far off, was it?

After all these trials, I decided one sure thing about DALL.E.  It works best with fewer parameters.  I put in two very simple requests for two different types of applique quilts.  First, I asked for an applique quilt with fruit on it.  These are the two pictures it gave me:

Not bad at all, and both are definitely quilts I could make.

Then –being the floral applique enthusiast that I am (because if you can’t grow flowers, at least you can stitch them) – it gave me these two renderings.

I really like both of those, and that basket of flowers may be in my future.

At the end of all this experimentation, I still believe A-I is like any other technology – it’s neither inherently good nor inherently evil.  It’s what’s done with it that matters.  It does bother me exactly how close it can get to the real thing.  One of my readers who is both a quilter and a tech person, and my daughter and son-in-law (both of whom are also tech-savy) relieved my fears a bit.  All of them told me most designers are putting some sort of imbedded code into their pictures on the internet.  For instance, a picture of a quilt pattern couldn’t be downloaded unless you paid for it.  Payment would override the code and allow you to use it.  I don’t think anyone programming AI would want to spend thousands of dollars just to peruse a lot of quilt patterns.

However, this also begs the question, what if those pictures were available on other websites?  For instance, a quilting magazine may feature the pattern.  Would it still have a code?  Could it still be “downloaded” into AI?  And what about folks like me who don’t sell patterns, but freely show pictures of their own quilts on blogs and other social media?  How protected are we from AI gobbling up our hard work?

I think, like most troubling fields in our society, the final decision will come down to the courts.  I believe there are workarounds with codes.  Remember when all our credit and debit cards were replaced with those with “chips” to prevent hackers from wreaking havoc on our accounts?  I honestly think those hackers said, “Challenge accepted.”  Two weeks after I received my first “chipped” card, my Visa card was hacked. I think the embedded codes will amount to the same thing. However, a lawsuit or threat of a lawsuit, is something everyone understands.  They take both time and money – mainly money. 

While I don’t believe AI is a huge threat to our quilting world yet, I do think it could be trouble in the future.  I’m interested to see how the courts rule on Andersen v. Stability AI et al and what it could mean to all artists.

Until Next Week, From My Studio to Yours,

Love and Stitches,

Sherri and Felix