Last month, visual artists across the internet raised their voices against machine learning-based image generators like Dall-E, MidJourney, Imagen, and Stable Diffusion — which generate images by sampling from an enormous dataset of artwork, photographs, and other images sourced from the internet.

This started on a large scale after the visual art portfolio site ArtStation featured an AI-generated piece on its explore page. Artists on the platform flooded its explore page with a graphic captioned “NO TO AI GENERATED IMAGES”.

Artists Are Upset

Artists across other platforms, including Loish and Mike Mignola, also expressed their rejection of the exploitative approach from these corporations; who train their software on copyrighted pieces of art without consent from and knowledge of their owners.

About a week ago, Getty Images published a statement announcing their legal proceedings in the High Court of Justice in London against Stability AI for copyright infringement. Getty Images claimed that Stability AI “unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images”.

Opt-In By Default

Google’s Imagen and Stability AI’s Stable Diffusion use datasets from LAION, which contain image-text pairs in the form of image URLs and their linked ALT descriptions. Imagen uses the LAION-400M dataset, while Stable Diffusion was trained on LAION-5B — a dataset that Stability AI helped fund the creation of.

Stability AI has announced that people can opt into or out of having their images used in the training of Stable Diffusion V3, by signing up to haveibeentrained.com and performing the steps necessary.

Many are still upset by the fact that Stability AI does not opt people out of having their images used by default.

As of now, ArtStation has updated their terms of service to disallow scraping and redistributing content that lives on their platform (which is the primary method for generating datasets). They have also prohibited the use of content marked ‘NoAI’ with ‘Generative AI Programs’ — a directive DeviantArt introduced to communicate such restriction. All posts on ArtStation now carry the ‘NoAI’ meta tag by default.

Why Is This Theft?

While some people are worried about ‘AI’ robots replacing them out of their jobs, many are concerned about the fact that these image generators are only as good as their datasets. Because there is a direct relationship between the datasets these programs ‘learn’ from and the output they are capable of generating, many artists feel that unauthorized use of their work to ‘train’ programs infringes their copyright.

On their website’s FAQ page, LAION says that their datasets “are simply indexes to the internet” — which are URLs to the original images, paired with their ALT text entries. This means that while LAION does not host images without their owners’ consent, they distribute a database containing URLs to access them — using which a user “must reconstruct the images data by downloading the subset they are interested in.”

This is rather interesting; because this might mean that LAION acknowledges the possibility — if not probability — of redistributing content without permission is an act of copyright infringement. It is worth considering whether distributing URLs and ALT text makes LAION the The Pirate Bay for machine learning programs, many of which profit off of software they train on these datasets.

It may also be worth considering what it means to train a program on pre-existing work. Is training an ML program simply a form of data compression, where the algorithm strips away unnecessary data and stores the rest in a new, reusable format?

Throughout this article and everywhere else on the web, you would notice that I tend to use the term ‘machine learning’ to address these programs, instead of the more widely used — and catchier — ‘artificial intelligence’. This is because I do not believe that these programs are intelligent. They are very predictable and do not make decisions that are not explicitly programmed. The reason you get seemingly random outputs from these programs is that they build their generations based on pseudo-randomly seeded noise.The output would also be identical if you were to generate a pair of images using identical prompts and seeds. Some machine learning programs let you access and reuse the seed.

The Tech Isn’t Evil, Corporations Are

Machine learning is a method for computers to ‘learn’ algorithms based on sample data. How could something so abstract possibly be responsible for enraging so many craftspeople?

You may have noticed already, but the flaw is not in the technology — it’s in the product.

Machine learning programs that learn from ethically-sourced datasets already exist. An example of this is Google’s Lyra, a speech codec trained “with thousands of hours of audio with speakers in over 70 languages using open-source audio libraries and then verifying the audio quality with expert and crowdsourced listeners.”

As Loish has pointed out, Harmonai’s Dance Diffusion — a machine learning-based music generator also released by Stability AI has trained exclusively on royalty-free music because “releasing a model trained on copyrighted data could potentially result in legal issues.”

Why Artists Are Upset With AI was originally published in Refuse on Medium, where people are continuing the conversation by highlighting and responding to this story.

I sent my first email newsletter.

It’s not the thing I was afraid of, though. Writing those newsletters is rather fun; I enjoy every part of the process. It also feels rewarding to have something finished and sent out to a group exactly every two weeks.

The challenge was finding people to send emails to. Neither my website nor my social media accounts drove enough traffic to find me willing subscribers for an email newsletter — a medium already infamous for privacy-invading, intrusive, spammy marketing. And it would not have made sense to work on weekly newsletters only to send them out to no one, or so few people that the likelihood of anyone reading my emails would be close to zero. I had no intention to half-ass them either; I would rather not have sent emails out at all.

So I did something I could never imagine.

I asked people to subscribe.

It might be the first thing people do when they make something new; but to me, it was scary. I had never asked someone to look at a piece of content — I could barely imagine asking someone to subscribe. My approach was always to leave a post, let people see it, and hope that they happen to be interested.

This time, however, I had to message people directly. I didn’t have a clever marketing scheme where I would offer people something of value for free — which would be free only if they agreed to receive it in an email. To make things trickier, nobody knew exactly what I would send them in the newsletter. I didn’t either. I just had a vague plan to experiment with small bits of hopefully-interesting content, and never be spammy.

To my surprise, it worked. It was simpler than I expected. I only reached out to people I knew could be interested; most were happy to subscribe.

The Plan (I Did Not Have)

Now that the worst was out of my way, it was time to build the newsletter. The fun part.

I had no systems planned out. I didn’t know how this newsletter was going to look. All I had was a general idea of it being not spammy, and a few text-files-worth of ideas.

I knew that I had my social accounts and a website to share updates from. The Refuse website had a huge and recent update at the time, so I had a handful of little details to share from that.

But it’s no better than following a social media account. I wanted the newsletter to carry substantial, but quickly-digestible snippets of content in multiple sections. Something that would be longer than a tweet and shorter than this blog post.

Another thing that was going to keep the emails interesting was the way it would look. Emails support very basic HTML styling, but that still offers a lot of potential from a visual perspective — at least as long as you do things right and work within specifications.

What My Emails Ended Up Looking Like

The first email ended up including these sections, and all future issues followed a very similar plan:

• A tiny note I write to remind readers that an actual person has written the email. It’s easy for newsletters to become too sales-y and focus entirely on products.
• The first email carried a small section explaining optics, which was continued in the next issue. This was an experiment with distributing article content in smaller chunks, across multiple emails. Not all issues of the newsletter have had this.

• I’ve had a few reoccurring sections which do not necessarily appear in every email. The earliest issues had design tips and updates from the website, and later issues had (mostly generative) experiments in visual art. Every issue until now has had a section for design wisdom.
• I make sure to draw a small cartoon at the end of every issue. It keeps the newsletter a little more interesting and light-hearted than it would be otherwise.

• I stylize every issue differently. Every single email has newly-drawn titles and a different colour palette from the previous issues. At the time of this article’s publishing, I’ve drawn titles for most issues by hand. It’s mail, after all. Here’s a browser version of issue #9, which has seamlessly-looping, animated titles.

How To Not Screw It Up

One thing I haven’t learned about writing email newsletters is what people prefer to see in their inboxes. At the time of publishing this article, the newsletter has less than 20 subscribers; and not all of them read every single email. I have no reliable data to learn from.

What I have learned though, is a few things that can improve or break emails.

You see, emails are a little similar to the conventional web; though a lot more similar to a text file. Because of its similarities to the web, emails sometimes look and perform very differently across different email clients and devices.

Here’s a small list:

• When possible, be a person. I include a personal note at the beginning of all my newsletter emails to remind my readers that I’m still the one writing to them. Remember that people’s inboxes aren’t ad-machines. Electronic mail is a medium for communication, and was built for people to talk to one another. This remains the primary reason people use it. Nobody signs up for an email account to get marketing content from businesses. Be respectful of this fact, and be grateful to people who care to subscribe to your newsletter.

• Some really popular email clients don’t support web fonts. It’s not a good idea to rely on a typeface hosted online, because many (if not most) recipients will see a different typeface. Make sure your emails look at least nearly as good with the fallback typeface as they do with the web-hosted font.
• Don’t use images to display text. Your email could be marked as spam. This is not a strict condition; but a high enough image-to-text ratio can alert email clients. Scammers sometimes try to hide phrases from spam filters by displaying them in images, which is why email services can mark your image-heavy emails as spam.
• Don’t rely on images to communicate with your readers. This is an accessibility issue for obvious reasons (some people use screen readers). But assuming you’re an a-hole who doesn’t care about accessibility anyway; email clients sometimes don’t show images to your recipients. It’s always nice to add alt text to images for when they don’t show up. As a small, extra step; I always add a copy of my drawn headings under them, but in plain text.
• Emails are text messages and you should treat them as such. This means that one of your primary goals should be to make sure that your emails render as quickly as possible. Always optimize your images to reduce their file size, and stick to reliable image formats. Most (of the popular) email clients don’t support embedded videos, so stick to GIF images for animations. When doing this, optimize your GIF images to minimize file size and maximize compatibility. A long render time does not make sense for emails.
• Make it easy for people to unsubscribe from receiving your emails. This is the right thing to do; but I will once again assume that you, the reader of this blog post, are an a-hole who doesn’t care about ethics, user experience, or being nice to people who are generous enough to agree to receive your emails. If my shaming hasn’t convinced you; many countries have anti-spam laws which require you to include a clear way for people to unsubscribe from receiving your emails in the future.
• Don’t fight for attention. In a world where people are conditioned to ignore attention-grabbing and sales-y phrases, it’s a really bad idea to make your emails look like every other piece of scammy garbage rotting in your inbox. Keep the subject line short, sweet, and accurate. People are tired of reading subject lines about yet another exciting thing; with the promise of being more exciting than every other email about an exciting thing in your inbox. I title my newsletters the following way: [An emoji] The Refuse Newsletter #[a number for the issue]. This subject line helps people understand what’s in the email, and who it’s from.
• This is an obvious one, but always send emails like these with consent from your recipients. You may want to go through privacy laws before you start misusing other people’s data for email marketing. Even if you don’t get into legal trouble, sending spam emails gets you moved to the — you guessed it right — spam folder.

I Just Sent My Tenth Newsletter Email. Here’s What I Learned From The Previous Nine. was originally published in Refuse on Medium, where people are continuing the conversation by highlighting and responding to this story.

At the time of publishing this article, Acid has over 200K opens. Which is pretty amazing — especially when it doesn’t quite have as many impressions yet.

That’s right; this Instagram filter has been working so well for people, that it consistently gets more opens than impressions. Which (I assume) means that it has a lot of regular users.

And despite this being a good thing for my AR effect, it had been something I didn’t like for a long time.

Really. I didn’t like that this one effect had been so much more successful than every other Instagram effect I had made — combined.

And the reason is that I was never going to release this effect. It was going to live as an experiment. This effect was a quick and fun experiment for me, while every other effect I had released previously, was more deliberate. I just happened to release Acid, because I happened to feel the need to release an AR effect, and be out of new ideas at the same time.

This reminded me that I was a bad judge; that I had wrongly optimized — wait, no — toned down other effects I had built, hoping to make them more accessible and neutral. As it turned out, Instagram (users) didn’t prefer neutrality as much as it preferred something that was extreme and experimental, even if it seemed to make very little sense from a design perspective.

And that made me feel like I never had the control I thought I did. That I was deliberately working against myself this whole time, while building all those other effects.

Things To Learn

As unexpected as it may have been, Acid’s great performance means that I get to have first-hand experience with what works.

Now — I can never truly know (unless you tell me) what makes people want to try this effect; then use it over and over. What I can do though, is think about what I did differently with this effect in comparison to others, make some guesses, and hope that I’m close enough.

This is why I think Acid has performed so well:

• Heavily Stylized: Acid is very stylized. It works with a few saturated colors, sharp contrast, and a lot of weird analog noise, which just happens to work right. People like art. It’s likely that being a creative effect is part of what makes people want to use Acid.

• It Just Looks Cool: I mean, that’s good enough of a reason for people to use an AR effect. That’s what almost all AR effects really do — make stuff look cool. An effect doesn’t necessarily have to serve another purpose. I think that Instagram’s AR platform has not even matured enough to allow it to a meaningful degree yet. Of course, you can attempt to build camera effects that do other stuff; but for now, you can only rely on Instagram’s AR platform to give you camera-toys.
• It Looks New: AR effects that manipulate and play with color are super common. But when it comes to working with color, you can always build something entirely new. And that’s probably why Acid works. It doesn’t look like most effects you would find someone using on Instagram. People like new things (though not too new. More about this in this great TEDx talk by Derek Thompson, on YouTube).
• It Obscures Your Face: Remember, how two bullet points ago, I pointed out that Acid is heavily stylized? It’s stylized so heavily, that it will partially obscure most things in the camera’s view. This includes faces; which might make it easier for people to take pictures of, or record themselves — should they hesitate otherwise.

• Not a Face Effect: Acid isn’t a face effect. Acid renders everything in the same style. It could be your face, a failing space rocket, a squirrel, or — well, you get the point. It’s a camera that renders things in black, red, yellow, and (mostly) green. And that makes it usable in more ways than a face effect would be.

• The Name Itself: The effect is called Acid, named after Acid Graphics; a style developed by a wave of contemporary designers — inspired by rave flyers, web-brutalism, and psychedelia; which this filter’s style is built upon. Maybe that’s one of the qualities of this AR effect that make people want to try it (although my hunch is that most users associate the name with the psychedelic drug and/or corrosive liquids).

How I Built It

I’m not going to get into the specifics of how the application I used to build the effect works, because that’s a different topic. It’s called Spark AR Studio, and has a bunch of online documentation, should you need help figuring out how to use it. If you’d still like to see a short and digestible guide on Spark AR Studio, let me know.

Building this AR effect was pretty straightforward. In fact, it was simple enough that I’m going to cover the process in bullet points (Spark AR has an updated guide on using color LUTs within Spark AR Studio, on their website):

• I created a rectangle that covered the full device screen.

• I extracted the camera texture by selecting the Camera object from the Scene panel, then clicking the plus (+) button next to Texture Extraction. This created a camera texture in the Patch Editor.

• I imported a patch called FastColorLUT into my Spark AR project, and fed the camera texture into that module. I fed a color LUT I created with Snapseed into the other node, and fed the output from the FastColorLUT module as a texture, into a new material with a Flat shader-type.

• What’s a LUT, you ask? Here’s an in-depth Wikipedia article about that. The kind of LUT I used, however, looks like this:

It’s an image, which contains a specific number of pixels. The color of each pixel in that image describes how a specific color will be recolored (or remapped to another color) upon processing an image through it. Every pixel in a LUT has a default color (a LUT containing default colors is called a neutral LUT), which is what image processing software compares your custom LUT to, before recoloring your image. As complex as that might sound, you won’t have to worry about any part of how it works.

• To create the LUT, I first imported a collage of people’s faces into Snapseed. I used a collage so I could check my LUT against different skin tones and a range of lighting conditions, without having to go through multiple revisions. I used Snapseed’s color-channel curves to build the look, and saved it as a preset within Snapseed. You can use Photoshop, or really anything else with RGB channel curves to do this. Heck, use DaVinci Resolve and you’ll still get the job done.

• I then imported a neutral LUT into Snapseed, applied the preset I had just saved, and exported the modified LUT as a PNG image. This was the LUT I fed into the FastColorLUT patch(making sure that image compression on that LUT was turned off); and that was all I really did to translate a Snapseed preset into a LUT, into an Instagram effect.

How I Built An AR Effect With 200K Opens was originally published in Refuse on Medium, where people are continuing the conversation by highlighting and responding to this story.

When was the last time you saw a shaded sphere in a logo, a photo covered with grunge overlays, or a rap song with just way too many triplets?

Trends come and go all the time. However, it can be painful for a lot of people to see a trend emerge, and eventually take over mainstream media. Everything starts to look the same, brands end up degrading their identities, artists lose quality and distinctiveness, and media platforms grow annoying; which pretty much explains why trends are looked down upon. It’s always a great idea to avoid using trends where they don’t belong. When it comes to design, trends provide no strategic value to the job, and make the products vulnerable to indifference from the audience. Plus, they just become annoying to look at. Although most trends revolve around some really powerful design elements; those elements might not always serve a purpose. Imagine looking up at a flower shop sign that’s styled in grunge. Doesn’t feel too right — right? Grunge-styled typography doesn’t communicate anything right about a flower shop.

Trends can kill your design for a lot of reasons:

• Trends usually offer no strategic value to design. Designing by a trend usually means that your job is inspired by what’s popular or trending, instead of what makes sense or complements the job.
• When trends emerge, they quickly take over media and industries. And if your brand chooses to follow a trend, it loses itself in a sea of all the other brands who have done the same.
• Every creator, business, and organization has a brand. And every brand — organized or unorganized — has a direction. Following a trend because it’s popular means that you’ll be likely to have your brand stray away from the direction that suits it best.
• Depending on what you do with them, trends can get annoying. Remember all the times you’ve seen Trajan get used on a movie poster. Traumatic.

It’s clear that design trends can easily ruin what could’ve been an acceptable work of design. So how should you handle trends? Boot them from your workflow altogether? Although chances are that a design trend will have nothing great to add to a composition; it’s still a design element, and still has something to offer where it fits. After all, trends are just design styles or compositions that happen to have grown really popular.

How Design Trends Originate

Design trends emerge for a lot of reasons. It could be anything from a product becoming really popular to a design style going viral. Every time something comes along that’s either really impressive, or gets a lot of attention for some reason; it’s susceptible to becoming a trend. Remember those annoying, 3D globes back from 2000s? People had them in almost every logo and every list paragraph because they were new; they were cool; and computers could now render them really fast. And oh; memes.

Should You Be Afraid?

Should you be afraid of design trends? Yes, and no. While trends can easily hurt your design; they’re still just design elements with their own style, place, and context to be used in. And if you try to confirm that — you might prove yourself wrong. There are just so many poorly-done examples of trending design styles, that the nicer ones get buried in the past.

Although if you look at the right places, you’ll find some really good examples of design styles that have now become trends. Trends aren’t unoriginal or bad by default. They have their own influences that make them them distinct. They just look really bad when they pass through a lot of unskilled hands. However, trends can perform just as well as any other design style once you understand how they work.

How To Work With Design Trends

There’s always a chance you might have to work with a trend. Here’s what you can do to keep your design on the right track:

• Know what you’re working with. Every visual style has an origin and a function; so a little research will always help. It’ll be much easier to make decisions once you have a clearer idea of what’s on your screen.
• Look up some good examples of the trend you’re trying to work with. Because trends go through and get worked with by so many people, chances are that you haven’t come across many pieces where a trending design style is used well. When working with a trend, look up pieces of design where it has been used well. Digging into old and iconic pieces of design can help you understand how a certain design works well.
• Design trends travel through a lot of eyes. And a lot of times, this may mean that a trending design style is perceived with different meanings across communities. It helps to research how a trend has evolved throughout its lifetime, and what it means to people — especially your target demographic.
• Acknowledge the reason you’re working with the trend. Does it have to do with function? Style? Aesthetic? Get the reason down, and do it well. Regardless of how a style is seen, nothing beats well-directed and well-executed design.
• Stick to the fundamentals. No matter what you’re designing, fundamentals stay the same. Due to the sheer volume of examples you’ll come across for any design trend, it’s easy to get influenced by any bad design practices you end up looking at. Although the internet can help you find inspiration, it’s really important to cut the noise and understand what influences are good or aren’t.
• If a trend seems hard to get right, tone it down — or ramp it up. No matter how common a design trend is, it might not work with everything. All design needs balance. If done right, toning things down and subduing those annoying elements can fix things. Alternatively, directing the entire design towards the style can add the consistency it needs.
• No matter what you design — do it right. One of the most common reasons design trends look bad, is that designers make them look crappy. Because so many people try their hands on trending design styles, most of the content you see is bound to be flawed. Always do your homework, and design things right.

What You Need To Watch Out For

Trends are just as brutal as they are annoying. Regardless of how well you’ve designed things, trends can do their own damage if you pick them for the wrong project. Here’s what you should watch out for:

• Design trends are a bad idea for things that revolve around a brand’s visual identity. When you show people something that’s too common, they don’t see it. Brands that are built well, will make you have strong, distinct, and positive feelings. And design trends are the perfect way to ruin that. Trends often feel common, thoughtless, and at least slightly annoying — exactly opposite of what a visual identity needs.
• Every design style — trend or not — has a history. Make sure you know what you’re working with really well; because what’s a cool work of design to you, might end up pushing your target audience away. After all, you wouldn’t advertise high-definition displays with a retro, 8-bit typeface; or, uh — issue a press release in Comic Sans.
• Of all the design styles that have existed throughout the history of mankind, not one fits all. It can be tempting to use trending design styles because they look really cool; or comfortable to give in to a trend because everyone else in the industry is doing it. No matter how safe it may feel, giving in to trends only takes away from design.
• Usually, things turn into trends when people really like them. And designers — regardless of how awesome we are — are people too. And like all other people, we have our own personal tastes and preferences. However, letting too much of that preference seep in through your fingers and onto the screen usually isn’t a good idea. Good strategy makes good design, and letting your (or your client’s) preference influence the design will usually ‘weaken’ it.

Originally published at refuse.ink.

Design Trends: Should You Avoid Them? was originally published in Refuse on Medium, where people are continuing the conversation by highlighting and responding to this story.