#bing image creator | Explore Tumblr Posts and Blogs

tangibletechnomancy · 6 months

Text

The (Personal) Is (Political)

~7 hours, Dall-E 3 via Bing Image Creator, generated under the Code of Ethics of Are We Art Yet?

Or, Dear Microsoft and OpenAI: Your Filters Can't Stop Me From Saying Things: An interactive exercise in why all art is political and game of Spot The Symbols

A rare piece I consider Fully Finished simply as a jpeg, though I may do something physical with it regardless. "Director commentary" below, but I strongly encourage you to go over this and analyze it yourself before clicking through, then see how much your reading aligns with my intent.

Elements I told the model to add and a brief (...or at least inexhaustive) overview of why:

Anime style and character figures - Frequently associated with commercial "low" art and consumer culture, in East Asia and the English-speaking world alike, albeit in different ways - justly or otherwise. There is frequently an element of racism to the denigration of anime styles in the west; nearly any American artist who has taken formal illustration classes can tell you a story of being told that anime style will only hinder them, that no one will hire them if they see anime, or even being graded more harshly and scrutinized for potential anime-esque elements if they like anime or imply that they may like anime - including just by being Asian and young. On the other hand, it is true that there is a commercial strategy of "slap an anime girl on it and it will sell". The passion fans feel for these characters is genuine - and it is very, very exploitable. In fact, this commercialization puts anime styles in particular in a very contentious position when it comes to AI discussions!

Dark-skinned boy with platinum and pink [and blue] hair - Racism and colorism! They're a thing, no matter how much the worst people in the world want you to think they're long over and "critical race theory" is the work of evil anti-American terrorists! I chose his appearance because I knew that unless I was incredibly lucky, I would have to fight with this model for multiple hours to get satisfactory results on this point in particular - and indeed I did. It was an interesting experience - what didn't surprise me was how much work it took me to get a skin color darker than medium-dark tan; what did surprise me was that the hair color was very difficult to get right. In anime art, for dark skin to be matched with light hair and eyes is common enough to be...pretty problematic. Bing Image Creator/Dall-E, on the other hand, swings completely in the opposite direction and struggles with the concept of giving dark-skinned characters any hair color OTHER than black, demanding pretty specific phrasing to get it right even 70% of the time. (I might cynically call this yet another illustration against the pervasive copy-paste myth...) There is also much to say about the hair texture and facial features - while I was pleased to see that more results than I expected gave me textured hair and/or box braids without me asking for it, those were still very much in the minority, and I never saw any deviation from the typical anime facial structures meant to illustrate Asian and white characters. Not even once!

Pink and blue color palette - Our subject is transgender. Bias self-check time: did you make that association as quickly as you would with a light-skinned character, or even Sylveon?

Long hair, cute clothes, lots of accessories - Styling while transmasc is a damned-if-you-do, damned-if-you-don't situation, doubly so if you're not white. In many locations, the medical establishment and mainstream attitude demands total conformity to the dominant culture's standard conventional masculinity, or else "revoking your man card" isn't just a joke meant to uphold the idea that men are "better" than women, but a very real threat. In many queer communities, especially online, transmascs are expected to always be cute femboys who love pink (while transfems are frequently degraded and seen as threats for being butch), and being Just Some Guy is viewed as inherently a sign of assimilationism at best and abusiveness at worst. It is an eternal tug-of-war where "cuteness" and ornamentation are both demanded and banned at the same time. Black and brown people are often hypermasculinized and denied the opportunity to even be "cute" in the first place, regardless of gender. Long hair and how gender is read into it is extremely culture-dependent; no matter what it means to you, if anything, the dominant culture wherever you are will read it as it likes.

Trophies and medals - For one, the trans sports Disk Horse has set feminism back by nearly 50 years; I'm barely a Real History-Remembering Adult and yet I clearly remember a time when the feminist claim about gender in sports was predominantly "hey, it's pretty fucked up that sports are segregated by sex rather than weight class or similar measures, especially when women's sports are usually paid much less and given weirdly oversexualized uniforms," but then a few loud living embodiments of turds in the punch bowl realized that might mean treating trans people fairly and now it's super common for self-proclaimed feminists - mostly white ones - to claim that the strongest woman will still never measure up to the weakest man and this is totally a feminist statement because they totally want to PROTECT women (with invasive medical screenings on girls as young as 12 to prove they're Really Women if they perform too well, of course). For two, Black and brown people are stereotyped as being innately more sporty, physically strong, and, again, Masculine(TM) than others, which frequently intersects with item 1...and if you think it only affects trans women, I am sorry my friend but it is so much worse and more extensive than you think.

Hearts - They mean many things. Love. Happiness. Cuteness. Social media engagement?

TikTok - A platform widely known and hated around these parts for its arcane and deeply regressive algorithm; I felt it deserved to be name/layout/logodropped for reasons that, if they're not clear already, should become so in the final paragraph.

Computers, cameras and cell phones - My initial specification was that one of the phones should be on Instagram and another on TikTok, which the model instead chose to interpret as putting a TikTok sticker on the laptop, but sure, okay. They're ubiquitous in the modern day, for better and for worse. For all the debate over whether phones and social media are Good For Us or Bad For Us, the fact of the matter is, they seem to be a net positive-to-neutral, whose impacts depend on the person - but they do still have major drawbacks. The internet is a platform for conspiracy theories and pseudoscience and dangerous hoaxes to spread farther than ever before. Social media culture leaves many people feeling like we're always being watched and every waking moment of our lives must be Perfect - and in some senses, we are always being watched these days. Digital privacy is eroding by the day, already being used to enforce all the most unjust laws on the books, which leads to-

Pigs - I wrote the prompt with the intention that it would just be a sticker on the laptop, but instead it chose to put them everywhere, and given that I wanted to make a somewhat stealthy statement about surveillance, especially of the marginalized...thanks for that, Dall-E! ;)

Alligators - A counter to the pigs; a short-lived antifascist symbol after...this.

Details I did not intend but love anyway:

The blue in the hair - I only prompted for platinum and pink in the hair, but the overall color palette description "bled" over here anyway, completing the trans flag, making it even more blatant, and thus even more effective as a bias self-check.

The Macbook - I only specified a laptop. Hilariously ironic, to me, that a service provided through Bing interpreted "laptop" as "Macbook" nearly every time. In my recent history, 22 out of 24 attempts show, specifically, a Macbook. Microsoft v. OpenAI divorce arc when? ;) But also, let us not forget Apple's role in the ever-worsening sanitization of the internet. A Macbook with a TikTok sticker (or, well, a Tiikok sticker - recognizable enough) - I can think of little more emblematic of one of the main things I was complaining about, and it was a happy accident. Or perhaps an unhappy one, considering what it may imply about Apple's grip on culture and communications.

Which brings me to my process:

Generated over ~7 hours with Dall-E 3 through Bing Image Creator - The most powerful free tool out there for txt2img these days, as well as a nightmare of filters and what may be the most disgustingly, cloyingly impersonal toxic positivity I've ever witnessed from a tool. It wants to be Art(TM), yet it wants to ban Politics(TM); two things which are very much incompatible - and so, I wanted to make A Controversial Statement using only the most unflaggable, innocuous elements imaginable, no matter how long it took.

All art is political. All life is political. All our "defaults" are cultural, and therefore political. Anything whatsoever can be a symbol.

If you want all art to be a substance-free "look at the pretty picture :)" - it doesn't matter how much you filter, buddy, you've got a big storm coming.

#ai art #generated art #dall-e 3 #bing image creator

233 notes · View notes

aismoker · 27 days

Text

How to Bing?

Part 2: Complex images

To go to part 1: click here

In my last post, I tried to explain the basics of creating images with the Bing Image Creator. In this part, we are going to dive into more complex images, mostly images with multiple people in them. This can be a bit tricky sometimes and as we are dealing with more complexity, the AI often does not exactly what you want.

As stated in the last post, I am no expert and I have not a clue how the algorithms work. There are probably plenty of other ways to make images, but this is the way how I make them. If you have tips or tricks, share them below.

STEP #1: Composition

For the examples I will stick to two persons, but you can also put more people in. Just know that more people doesn't necessarily make the image better, especially when you make it more complex. For example, "Five Swedish bodybuilders sitting in the snow. they are sweating", will probably work just fine (although the AI probably has some problems with counting to five), but if you put more details in to diversify them, it will get messy. I think it has to do with the fact that the generator sees the five Swedish guys as one entity, but that can be just a weird idea from my side.

As with images with only one person, the first step is to think about the composition. Since you have now multiple persons, the phrasing is more important here. In the prompt "A man stands in the forest.", you can simply replace "a man" with "two men" and you will get a nice picture of two similar looking men standing in a forest (examples 1 and 2 below). If you want to diversify, you should, however, approach it differently. Instead of "two men", you should write "a X man and a Y man", in which the X and Y can be different specifications, as in example 3 and 4 below (I have no clue where the medieval clothes came from in example 4, but hey, why not?). The differences, however, can be very small. It is, however handy that you can refer in the rest of your prompt to both of them separately. It is, of course also possible to replace the word man with something else, like in example 5. A third possibility is to make one of the two people more dominant, as in example 6. The focus will usually be on him and the other person is basically a prop.

"Two men stand in a forest."

"A Norwegian and an Italian man stand in the forest."

"A 30 yo man and a 25 yo man stand in a forest."

"A 30 yo bouncer and a 25 yo security guard stand in a forest."

"A 30 yo man stands in a forest. A 25 yo man stands besides him."

STEP #2.1: The duo

We start with the duo (example 1 and 2 above). The best prompts are those, in which they do both the same thing and wear the same attire (obviously). This kind of prompt is also very useable if you want to make an image with more than two people. It is possible to diversify with "One of them... The other", but it makes the generator struggle and he starts to make mistakes, simply because he doesn't understand what attribute belong to one person and what to the other. You can try it out and there are sometimes very good results, although it is usually not exactly what you asked for. But the more complex it gets, the lower the success rate.

"Two 26 yo bikers stand in a dark forest. they wear black biker jackets. they wave."

"Two 26 yo bikers stand in a dark forest. they wear black biker jackets. One of them waves. The other one smokes a cigar."

"Two 26 yo bikers stand in a dark forest. They wear black biker jackets. One of them waves. The other one smokes a cigar."

"Two 26 yo bikers stand in a dark forest. They wear black biker jackets. One of them has a shaved head and wavers. The other one has curly hair and smokes a cigar."

"Two 26 yo bikers stand in a dark forest. One of them has a shaved head and he wears a green leather biker jacket and he smokes a cigarette and waves. The other one wears a black shirt and has curly hair and smokes a cigar."

As you can see, there is little wrong with example 1. It is also a quite simple prompt. Examples 2-4 are already a bit more complicated. You see already a small derivation from the prompt in example 3, as the cigar-smoking guy shouldn't be smoking. They are still, however quite okay. The last prompt is, however very complicated. To make it understandable for the generator, I linked everything with "and" together, but to no avail, It is one of the few pictures in which actually one of them has a cigarette in his mouth and the other one a cigar, but I had hoped, that the guy on the right side would just wear a black shirt, without a jacket. It can be that the jacket appears, because the guy is described as a "biker." More problematic is, that that same guy has seven fingers. Example 6 is less disturbing and is in my opinion perfectly useable, but it is not following the prompt.

STEP #2.2: 1 + 1 = 2

A different method is to create a prompt with two individuals, like examples 1.3-5 ("A and B do something"). In this way you can define A and B already in the first line of the prompt. You can then also add attributes by saying "A does/wears/has something. B does/wears/has something". This method is not flawless, but works in my opinion better than the duo method.

"A 35 yo bald bouncer and a 32 yo business man stand in a forest. They kiss."

"A 35 yo bald bouncer and a 32 yo business man stand in a forest. They kiss. The bouncer wears a shiny black pvc puffer jacket. The business man wears a fedora."

"A 35 yo bald bouncer and a 32 yo business man stand in a forest at night. They kiss. The bouncer wears a shiny black pvc puffer jacket. The business man wears a fedora. The bouncer holds a cigar in his hand."

As you may notice, the latter prompts are quite complex, but there is no contamination between the attributes of the two men, except in the third example, in which the business man suddenly wears a shiny black jacket, but I like it, so I call it a happy accident

STEP #2.3: prominent and prop

With this method you can create an image with 1 person is more prominent and the other person is more like a prop, like in example 1.6, although the difference between prominent and prop fade quite quickly. I have mixed feelings about this method. Sometimes I get seriously amazing images and sometimes they are horrible. It works quite well in my experience if you keep it simple, without a defined background and all.

1. & 2. "A 30 yo baldbouncer leans against a tree in a forest at night. He smokes a cigarette. He wears a biker suit. A 25 yo business man approaches the bouncer."

3. "A 30 yo bald bouncer leans against a tree in a forest at night. He smokes a cigarette. He wears a biker suit. A 25 yo business man cleans the boots of the bouncer."

4. & 5. "A 30 yo bald bouncer leans against a tree in a forest at night. He smokes a cigarette. He wears a biker suit. He hugs a 25 yo business man."

6.-8. "A 30 yo bald bouncer leans against a tree in a forest at night. He holds a cigarette. He wears a biker suit. He exhales a cloud of smoke. He kisses a 25 yo business man."

Examples 1-3 are quite simple. As you can see, the focus is on the biker and the business man is featured less prominent. Examples 4-5 are of prompt that I and the generator struggle with quite often. In both examples the biker should be the one who is hugging the busines man, but in example 4 it is the other way around. I noticed that words like "hug", "embrace", etc. confuses the AI, just like phrase like "A stands in front/behind B" and that the generator often switches the persons around. Just give it a few extra tries with the same prompt and the generator will eventually place the people in the right spots. If that doesn't happen, you can try to switch the persons in your prompt around, so "the busoness man hugs the biker". In examples 6-8 you can also see that the generator is struggling. Example 6 looks good at the first glance, but if you look closer, you can see that the hand on the tree, or whatever it might be, is floating around without an arm. Example 7 os nice, but the persons are switched around. When this happens, just try again and eventually you will holefully get the desired image, or something close to it, like example 8. I am quite satisfied with that outcome, but the smoke could have be done better, but maybe I am to critical.

STEP #3: Combining methods

Of course, you can also combine the three methods I explained above. Sometimes it works, sometimes it doesn't, just try it out.

"A 35 yo bald bouncer and two 25 yo scally lads stand in a forest. The bouncer wears a green bomber jacket. The scallies wear shiny black tracksuits."

"A 35 yo bald bouncer and two 25 yo scally lads stand in a forest. The bouncer wears a green bomber jacket. The bouncer smokes a cigar. The scallies wear shiny black tracksuits. The scallies kiss."

"A 35 yo bald bouncer and two 25 yo scally lads stand in a forest. The bouncer wears a green bomber jacket. The bouncer smokes a cigar. The scallies wear shiny black tracksuits. The scallies kiss each other."

"A 35 yo bald bouncer and two 25 yo scally lads stand in a forest. The bouncer wears a green bomber jacket. The bouncer smokes a cigar. The scallies wear shiny black tracksuits. One scally kisses the other scally."

"Two 25 yo scally lads stand in a forest. They wear shiny black tracksuits. They kiss. Behind them stands a 35 yo bald bouncer. The bouncer wears a green bomber jacket. The bouncer smokes a cigar."

The first image is a very easy one: 1 bouncer in a bomber jacket and 2 scallies in shiny gear, doing nothing. This is a combination of 1 + 1 = 2 composition and the duo. I then wanted to make the scallies kiss and make the bouncer smoke a cigar. In examples 2 and 3 my description wasn't clear enough, so I tried it differently. Unfortunately I didn't get the desired result (examples 4 and 5). I therefore decided to use the duo in combination with the prominent and prop. I am still not really satisfied and I should tweak the prompt further, but at least the guys all do what I want them to do.

As with most things in life, you need to practice with prompts and after a while you will get a feeling for it.

BONUS: Before-and-after pictures

Since you guys seem to like the whole transformation shebang, I have one more composition that I want to share with you at the moment. I didn't come up with it, but I found it on the page of @nolecub, an amazing story writer and magnificent AI-image creator. It is due to his posts, that I decided to try the AI-image generator.

The basics of a before-and-after picture is that the screen gets divided into a "before" and an "after". Of course you can make two separate images, but the problem with Bing is, that you cannot make one character (person, animal, or object) and put it in different images. So, every "guy with black hair" looks totally different from the next one. Of course, there is a change that he creates two guys who look a lot like each other, but that can take you several tries. With the before-and-after prompt, you stay in the same picture and therefore, the generator will usually get what you mean (sometimes it will not, of course). One of the problems with this prompt is, that you basically have to describe two images, which can cause some problems with the character-limit. So, before you start, make sure that you know what you want and focus on that.

The prompt is quite simple:

"Before-and-after picture.

Before: [insert content]

After: [inset content]"

You can leave the the first line out, but then the change that the before-and-after blur into each other is larger, It can cause very cool effects, though!

Here are some examples:

"Before-and-after photo. Before: A 25 yo nerdy guy stands in a city. After: The same guy, but he is now a redneck. He stands in a forest."

"Before-and-after photo. Before: A 25 yo gamer. he holds an unlit cigar in his hand. He has curly hair. After: The same guy, but now he smokes the cigar. He wears a leather jacket. He has a large beard and he has shaved his head shiny bald. He has gained muscles."

"Before-and-after photo. Before: A 25 yo average guy. After: The same guy, but now a year later. He has joined the military. He smokes a cigarette. He has a military haircut."

"Before-and-after photo. Before: Portrait of a 25 yo man. He stands in a city. It is summer. After: The same guy, but now he is a bodybuilder. He stands in a snowy forest. He has tattoos on his arms. He smokes a cigar. Tendrils of smoke surround him."

As you can see, I usually start the "after" with "the same guys, but now..." When it comes to body changes like muscle gain, I usually use words like "he gained muscles", instead of "he is muscular". These kind of images are quite difficult to grasp for the generator and it happens often, that the desciption of the "after" seeps into the "before"-part. With phrases like these, I always hope to help the generator a bit, but I am not sure if it really works.

That's all for now. I hope this helps some of you guys out in making some nice images. Dont'forget to share them!

If you have any questions, or if you have found a better way to create a certain kind of image, let me know!

#ai art #ai artwork #ai tips #bing image creator

84 notes · View notes