Artificial intelligence knitted it!

What is your hobby? Surely you have something that brings you joy. It is said that a hobby should be something “useless,” meaning something that does not directly create a functional or usable product. If we create something useful, according to psychological teachings, we are actually constantly working. The type of work changes, but we always have in the back of our minds not creativity, but the need for a functional result.

Words by Zuzana Grúňová

I don’t know how true this is, I haven’t looked into it closely. I told myself that if it brings me joy, that is the definition of a hobby where one can relax. Whether the result is a decorative dessert, a hat, or a model for an electric railway.

So, I knit. At first, small things, ordinary, simple. Gradually, I began to discover the charm of various portals that bring together similarly “afflicted” people. In addition to thousands of free tutorials and millions of different types of patterns, there is also something that could be called micro-entrepreneurship.

An enthusiast creates a tutorial, shares it on a portal for a small fee as a digital file with instructions and recommendations. Some more complex tutorials are so well-crafted that they require the involvement of others. They do testing – they knit the item according to the tutorial and uncover errors, typos, or ambiguities. There is also the release of tutorials in parts over a longer period, where the community knits the item piece by piece (CAL Crochet-A-Long, KAL Knit-A-Long). And many other games. You are probably smiling now. Every community of practically anything is like that… a bit strange, perhaps, but believe me, the people in it enjoy it.

Great, and where is the promised AI? I’m getting to that. (Micro)entrepreneurship is so interesting that on major portals like Ravelry, even big magazines, or individual designers publish their patterns. And these are not “ordinary” sweaters, but literally works of art. Dead serious. Try looking up designs by Alice Starmore, Amalie Røge Hove, Erika Knight, or the magically colorful Kaffe Fassett. Among them, perhaps hundreds of thousands of others create and offer their patterns online for download. They are less known, but for a perfect pattern for my son’s hat and gloves, I thank them and was happy to support them with a small amount for their pattern.

Now imagine that this parallel micro-world of knitting and crocheting is attacked by a virus. It destroys trust, and thus the small financial support that selling patterns and tutorials generates for many.

What am I talking about? Imagine you are a fraudster. You generate some hyper-realistic photos of children in adorable knitted items using AI. Or amigurumi, small crocheted figures and toys. Many of these images will be practically indistinguishable from real creations, even for experienced knitters and crocheters. You then use an AI language tool to generate the content of the tutorial. The entire text, notes, instructions… everything. In the data files on which the models were trained, these data are also included. Millions of samples, freely available in blogs, on yarn manufacturers’ websites… The resulting tutorial looks trustworthy at the first glance, it contains everything it should. From the introduction, description of the yarn, needle sizes to the entire process.

Knitting is actually something like mathematics. The stitches on the needle can be manipulated, but it’s about the number – some are decreased, others are cast on, added in various ways. In the knitted fabric, we create stitches, overlap some parts, and create not only the structure but also the overall shape. So that the gloves fit the hand, the hat fits the head, and the knitted top elegantly hugs the shape of the body. Mathematics can be nicely simulated.

Now, a fraudster sells such a pattern. Many people pay for it, download it… and find out that the result is nonsense, it is unworkable. Or it doesn’t look anything like the enticing picture. Most of them probably won’t pursue the damage worth a few euros. Some may give up the hobby and no longer trust the entire portal. Others consider it their fault and never return to knitting, with their soul wounded by another disappointment. I can’t do anything, nothing works for me. And that’s a shame.

It’s not the fault of AI, of course. It’s just a sad reflection that we, humans, have a tendency to misuse every technology.

The first time I encountered this ugly phenomenon was on social media, where many shared attractive, colorful photos of knitted and crocheted creations, generated images using AI. Some admitted that it was AI and shared them for inspiration and to see if a functional pattern, a real creation, could actually be derived from them. Others, unfortunately, stubbornly spread the generated images as “their works” and even tried to sell finished “real” products. Over a few years, an atmosphere of disgust and suspicion developed. Several guides on how to recognize these generated images have also emerged – most of them, anyway. Because AI models are becoming more and more perfect.

In March 2023, I found a different approach – to try to use AI. To ask it to create a pattern and then test it. This was done by a creator from the YouTube channel Future Fiber: I asked an AI to generate a knitting pattern (& made it!) https://www.youtube.com/watch?v=YEUyCweWtDc

She followed the instructions exactly to create a cabled neck warmer. However, the process did not create any easily visible, repeating pattern, nor did it form the desired three-dimensional parts of the knit as is usual with this type of pattern. Thus, the generated code uses the names and terms, but the resulting knit does not correspond to them. It lacks a discernible pattern that a knitter would practically learn by repetition when following a human designer’s pattern and be able to predict which type of stitch to use and also spot mistakes. The pattern generated by Chat GPT was complex, inconsistent, and non-intuitive. She eventually published this pattern along with a description of the result on the portal. https://www.ravelry.com/patterns/library/cozy-and-textured-cowl

The comments from the creator and her audience are telling:

“I actually made a video that is mostly about experimenting with ChatGPT and what it could do for knitting, but I think we are also starting to deal with a discussion about ethics and consent and how it could affect knitting designers.”

“Fascinating and at the same time disturbing. So many things are happening around AI – I never thought about how it would affect artists and their work. Wow!!”

“We knitters and crocheters work hard on our craft and products… and it’s more than a little disturbing to be overtaken by fake products… and if the general public believes they are real and achievable.”

“Thank you for sharing. I hate AI even more knowing it has penetrated my knitting world.”

I tried it too. I had Copilot create a pattern – here it is:

“I’d be happy to help! Here is a simple lace knitting pattern for a 20-stitch sample:

Materials:

  • Yarn of your choice

  • Needles appropriate for your yarn

Abbreviations:

  • k: knit

  • p: purl

  • yo: yarn over

  • k2tog: knit two together

  • ssk: slip, slip, knit

Pattern:

  1. Cast on 20 stitches.

  2. Row 1 (RS): k2, yo, ssk, k2tog, yo, k2; repeat from * to end.

  3. Row 2 (WS): purl all stitches.

  4. Row 3 (RS): k2, ssk, yo, k2, yo, k2tog; repeat from * to end.

  5. Row 4 (WS): purl all stitches.

  6. Row 5 (RS): k2, yo, k2tog, ssk, yo, k2; repeat from * to end.

  7. Row 6 (WS): purl all stitches.

  8. Row 7 (RS): k2, k2, yo, k2tog, ssk, yo; repeat from * to end.

  9. Row 8 (WS): purl all stitches.

Repeat Rows 1-8 until your swatch reaches the desired length.

Bind off all stitches.“

I tried to go even further. I copied the text and used it as a prompt for a TXT2IMG AI tool.

Fig. 1. Image generated by artificial intelligence as a model

And finally, I knitted the pattern EXACTLY according to the instructions. I was surprised that at least it maintained the same number of 20 stitches after knitting a row. So it seems that the AI either “understood” or just by chance maintained the number of stitches decreased and cast on in each row. The pattern is nonsensically random, does not repeat, and does not create what the human eye considers an aesthetic structure in knitting. It is a random arrangement of holes. It also has nothing to do with the generated image above. While visually attractive, it is absolutely unrealistic to knit – its parts look like they are woven, knotted, crocheted (and it contains 34 stitches).

Fig. 2 Actual result

Conclusion? As an interesting experiment, it is fun to generate inspirations for knits or other textile processes. However, general models cannot create a real “model” of knitting according to a text prompt, nor generate a truly functional pattern with an aesthetic structure. For now.

But I wouldn’t completely dismiss AI as a creator of knits. In industrial knitting, it is an extremely useful assistant. However, they must be narrowly specialized AIs, directly trained on large datasets of knits and patterns. They also have a limited, precisely defined set of types of manipulations with knitting stitches. They then try to identify these on the finished product and reconstruct the pattern backwards. We all wear industrially knitted patterns almost every day – even a regular T-shirt is a knit, a fabric made from very thin yarn. And industrial knits sometimes have many other advantages. In production, it is possible to program knitting machines to produce entire pieces of clothing so that they do not need to be sewn together. There is no waste, no extra work, and seamless items are more comfortable to wear and less likely to deform.

Lesson? Let’s look for meaningful tasks for AI, and with an ethically correct approach, it will be an enrichment – even in knitting.

Sources:

future fiber : I asked an AI to generate a knitting pattern (& made it!) 

_TAGS

_PARTNERS