Flux 2 democratizes image generation

Last year, I used Flux schnell to generate a banner image for my Medium article (link here). The quality was quite good for an offline model, but it took hours to generate even a small picture on my laptop, so I never used it again. After all, Nano Banana by Alphabet is free and fast, and on top of that of superior quality (you can read my other Medium article about Nano Banana here).

Still, it is enticing to generate images offline, where you have more control on parameters like aspect ratio (AR), resolution, guidance scale, etc. So I tried out Flux 2 klein, the successor of Flux, and was positively surprised: it is way faster than Flux (1) and the quality is even better. 😍

Given the ease of use and the good quality and controllability, the users have full freedom in creation and can even generate images which look surreal but feel more real than the reality, which I will show you next. But first let’s warm up a bit and see what Flux 2 klein can do technically.

Unusual AR is easy peasy

You know, a banner image has to be wide. I forgot the exact width and height I used back then, so I simply let Flux 2 generate the banner anew with an AR of 2 (double as wide as high), and then of 3 (triple as wide as high).

I also forgot with what prompt I generated the old banner, but a look at it easily infers that the prompt would be like this:

a robot working on a blog article at the computer

Tada:

AR 2:1

AR 3:1

I find the eye part of the robots fairly cool. What do you think?

BTW: the width and height should be multiples of 32, recommend are 1024x1024 or 1024x768. So only use an unusual AR if necessary.

How about more wild prompts?

Just for fun, I asked Pi the chatbot for some wild example prompts and here is one:

A neon-lit raccoon DJ spinning records on Mars, surrounded by floating jellyfish wearing tiny sunglasses

Ok, maybe not as wild as the many AI images you can find on the Internet, but cute it is:

Quite impressive for a small offline model, isn't it? I would go on Mars if there are a raccoon DJ and sweet little jellyfish with tiny sunglasses awaiting for me 😁

Phantasy that feels real

So far, Flux 2 has only generated phantasy pictures which are obviously not real. For an AI model without human-like experience and understanding, it’s even easier to generate unrealistic pictures like a horse riding an astronaut. 🤭 Possibly annoyed by mass of such AI images (AI slops?), many try to make AI generated images look more real.

As usually, I tried the other direction. 🙃

Also phantasies can have values and even feel real. As you know, many identify themselves more with fairytale figures like Cinderella or Snow White than with real world figures. And after the invention of photography, artists like Picasso or van Gogh made paintings which obviously don’t look real, but somehow feel real and resonate well with their fans.

You can find more such examples in the art or literature, not only in the classics, but also in pop songs like Have You Ever Seen the Rain by Creedence Clearwater Revival.

“… There’s a calm before the storm .. When it’s over, .. It’ll rain on a sunny day … shining down like water … Sun is cold and rain is hard …”

In more tropical regions, it can indeed rain on a sunny day, but rather short and light, not “shining down like water”, never “sun is cold and rain is hard”. Still, the lyric feels real, more real than the reality (and yeah, I saw it with my inner eyes).

I’m not sure if CCR also saw it this way, but here is what Flux 2 generated:

How did I do it?

Since Flux 2 is technically capable of generating images the way you want, the only thing you have to do is to find the prompt that leads to the desired result. A tip is that if the generated image is already close to your concept, then instead of changing the prompt, you can try changing the seed to see if another version of the prompted image would look better.

Here is the prompt I used:

A giant, human-dwarfing futuristic skyscraper of steel and glass being lit by a glaring, eye-hurting sun, icy rain pouring down from the crystal-clear blue sky, resembling a down-rushing waterfall.

Why did I prompt with “a giant, human-dwarfing futuristic skyscraper of steel and glass”? While tech and art are actually complimentary and go hand in hand in the production process, they do represent two distinct, if not opposite, ideals. Tech is the result of sophisticated teamwork, you need to collect capital, then you have to coordinate resources, make sure that everyone is equipped with the right tool, deployed in the right place at the right time. Often, the team is so large and complex that you lose the oversight, and feel small and powerless in front of the colossal monument of high tech, especially when you are out of the team.

So I imagine, when the artist, maybe CCR or someone else, maybe Esther, was again turned down by the agency, and all his ambition, talent and self-confidence was reduced to “just another number” in the long list of job applications lying around in the agency, when he went out of the agency and turned back with the faint hope that someone maybe changed the mind and came out to get him in, he saw that top modern agency building, shining in the sun which hurts his eyes, then suddenly rain came down “like water”.

Of course, that could happen to any of us, not just artists, that’s also why this song is so popular. I imagined in specific an artist in this scene, not only because CCR’ is credited with this song, but also because artists, but also sportsmen and -women, carry the other ideal, namely the power of individual: creativity, strength, speed, independence, freedom, free from capital and other constraints. That’s why we celebrate artists and sportsmen as solo stars and heroes although we know they are ultimately enabled by the teamwork. We know how important high tech is for the success of a concert, like how important a pair of good running shoes are for the new marathon record. Still, we focus on the individual artists and sportsmen as the front face. The reason for our desire to cheer an idol is because while recognizing the power of teamwork and cooperation, we also desire individual development, we want to become free and independent. Sometimes we even go so far and resent capital, globalization, or sort of things which make each of us look so small and sometimes put us into precarious or insecure states. So there is steadily this tension between the two ideals, collective achievement and individual development.

Flux 2 has no idea of the two competing ideals, of course, but somehow it magically added a few buildings in shadow to the picture and made it look even more suppressive and so better represent the feelings conveyed in the lyrics. I find it amazing. 🔮

By changing the seed from 0 to 42, I actually got Flux 2 to generate the image exactly the way as I imagined, namely with one single giant building. It also looks impressive, but after some considerations I came to conclude that Flux’s variant with more buildings in the shadows, with its stark contrast between dark and bright, actually better reflects the lyrics.

Here is the image which more resembles my original imagination:

It also looks good, but lacks a bit contrast which makes the CCR song so strong. Ok, it also contrasted sun with rain, but an additional contrast between dark and bright may appeal to the audience more, I thought. What do you think? Do you like this one or the previous one with more contrast better?

As you can see, changing the seed can alter quite a bit of nuance and sometimes add interesting features. There is no rule which seed number is the best, you have just to experiment around. Luckily, Flux 2 is fast, so it’s easy to generate several images to choose from.

More important is of course the prompt. Even a slight change of the prompt could change the image substantially. Originally I didn’t add the short phrase “resembling a down-rushing waterfall”, then the generated images would look like this way:

Now the rain looks even more real. But the purpose here is not to depict some real weather phenomena, but a feeling, as conveyed in the CCR song, for which the rain here would look too light, even a bit delightful. What do you think?

Another (counter-)example

A picture obviously looking unreal but somehow feeling real doesn’t always have to be rain shining down on a sunny day, it could also be the other way round, like, on a rainy or snowy day, the protagonists are totally immersed in warm golden sunshine, like this one:

The prompt I used:

A hansome young man and a pretty young woman with beautiful short brown hairs hugging each other in love. They are dressed with simple and light clothes and immersed in a warm golden light which covers them like a fog such that only the silhouette is distinguishable. It is snowing on the street in the background, which fades into indistinguishable by the intense golden light, no snow can touch the loving young couple.

I prompted the woman to have beautiful short brown hairs so that she resembles Della in O. Henry’s The Gift of the Magi, in which Della cut her beautiful hair to buy a worthy Christmas gift, a platinum fob chain, for her beloved Jim while Jim at the same time sold his precious watch to buy Della combs she always desired.

This picture obviously doesn’t look real, but I try to convey the feelings from the story: the outer world is cold and snowy, but the inner world of the loving couple is filled with warmth like a golden light which makes them untouchable to the snow and the coldness from outside. For it not to look too contra-physics, I also let Flux 2 to only depict the couple as a silhouette within the golden light, as something you maybe once saw in a forgotten dream, but Flux 2 decided to picture them more clearly, with half-closed eyes, and all the snows changing into light dots in front of them. Somehow it even looks better than my idea with the silhouette. 😛

Here you can also see how much could a slight change of the prompt alter the image generation. Without the silhouette phrase, the picture would look this way:

The couple smiles a bit too brightly so that it looks like a commercial, and it lacks that dreamy effect which renders it as a love story from the distant past as it is. Indeed, The Gift of the Magi, published in 1905, depicted a world so different from ours that you would not even recognize it as a love story and rather classify it as an irony as you can find on Wikipedia.

I actively searched this story to use it as a counter example to the rain on sunny day example above because I first heard about it in school, maybe at college, in China. The teacher, a young woman at that time, introduced this story as a love story and was visibly touched by the twists and turns. It’s a bit amusing that when I looked it up on Wikipedia, I read that the story was characterized as “cosmic irony”. I dare to say that my Chinese teacher may have better understood the story than the English-speaking author on Wikipedia. Because China was at that time not long after the so-called Reform and Opening, when the working conditions more resembled that in the story, set in USA in 1905: payment was generally poor and insecure, but people were full of hope, in China because of the end of the long isolation, in USA for reasons unknown to me (but I can see a lot of hope in O. Henry’s story). We know that understanding is often based on similar experiences, so I dare to say that Chinese in the eighties possibly better understood Americans in 1905 than Chinese born after 2000 would understand Chinese living in the eighties or Americans at the present (Wikipedia was founded in 2001) would understand Americans at the beginning of the last century.

Somehow, Wikipedia is also right to see the irony in this story. This irony would be more salient if you consider what came after 1905: WWI, The Great Depression, WWII… A young couple of 22 who lost one third of their income in 1905 would experience quite some hardship during the subsequent events when they grew older. When you look at it with hindsight, from this God’s perspective, you may well view it as something ironic, if not futile.

But if you look at it with the eyes of those living and experiencing it, if you look at it with the eyes of Della, who, in her last moment, having the most precious moments in her life passing by before her inner eyes in a flash back, you may well see, immersed in a golden light, there is:

Instead of obeying, Jim tumbled down on the couch and put his hands under the back of his head and smiled.

“Dell,” said he, “let’s put our Christmas presents away and keep ’em a while. They’re too nice to use just at present. I sold the watch to get the money to buy your combs. And now suppose you put the chops on.”

Final thought

While AI (and other high tech) and art can be understood as two opposite ideal, the tension between AI and art is rather a concept than an unchangable fact, I dare to say. In practice, AI (and other technics) are useful tools for artists, but also for anyone else like you and me: as you can see, how easily did I generate the pictures with Flux 2. So it was really no exaggeration to say that Flux 2 democratized image generation, everyone can use Flux 2 to show the world what they see with their inner eyes. On the other hand, a concept can also co-determine the fact, especially in teamworks. Many teams break up not because it is physically impossible to work together, but because there is no working concept to divide revenue and share risk. I won’t discuss the concept in this article, it would be too long. And somehow I’m optimistic that others will figure it out anyway.