Real use cases for stable diffusion

[Ehplodor]

Out of the true simple joy of being able to generate images from a few well chosen words, rivaling with the best artists in the world, I think a very important subject is "what is this really useful for ?"

• Video games / CGI

  • "generating portraits for a game mod of Age of Empires 3 Definitive edition" (TY @matrix4767)
  • tileable textures (TY @horribleCodes)
  • Tile finalization (TY @hydropix)

• product and architecture design

  • sketch2img (TY @horribleCodes)

• marketing (TY @Chilluminati91 @hydropix)

  • faster iteration rate
  • improved artistic direction overall
  • AR corn wallpaper https://twitter.com/Warvito/status/1570691960792580096?s=20&t=myRuGf4EPaX2VzZ27RHXGA

• Security

  • Anonymization of faces / persons / scene in photographs for social networks and other image-based communication

Sciences
- fake MRI dataset :-/ https://twitter.com/Warvito/status/1570691960792580096?s=20&t=myRuGf4EPaX2VzZ27RHXGA

I'll begin, then expand, this list everytime I find real use cases for this.

[matrix4767]

Does generating portraits for a game mod of Age of Empires 3 Definitive edition count as a real use case? I would think so.

[Ehplodor]

of course ! TY !

[JPhilipp]

In a sandbox universe where people create their worlds, it's nice to be able to have text-to-image as well as text-to-3d. You can then use speech recognition via e.g. Google Cloud, have NPCs converse via GPT-3, and build anything around you which you can dream of. I'm currently trying to build a tool set around this in Unity.

[ghost]

As an artist, I can arguably tell you that this kind of tech based around NLP is bringing an industrial revolution and the use cases are too many to count, mainly because the only limit we will encounter will be our imagination

[Ehplodor]

To the point. Comments until now are mainly focused on game and VR content generation. Is that the revolution you are talking about ? Or something bigger ?

[ghost]

Like it was said in the StabilityAI keynote a couple of days ago, these kinds of leaps democratize the ability to communicate in new levels. Studying and mastering complex tools wont be needed anymore as the focus is turning towards natural language, a knowledge barrier as low as it can get with current technology. What the printing press was to writing, what photography was to painting, and now what AI is turning out to be to, well, any media creation. This is just beginning of course, as you need some technical knowledge to set it up. But as we train AI to understand our language more and more, we are aiming to forget about having to resolve many if not all sofware issues and be able to just ask for stuff, whatever that is. Videos? Stories? Experiences? Code? Knowledge? This paradigm crosses every field and its growth is exponential. Take a look at things like https://github.com/wpeebles/G.pt , we're teaching AI to teach and program itself.
So yeah, thinking about the implications is exciting and terrifying at the same time.

[Ehplodor]

I'll have a look. Thank you

[hydropix]

Many things in the game "Have a nice death",like the fire background here https://youtu.be/AYx53xGAqiQ?t=4233 (thanks tile feature). And also sprites finalization with img2img and a very low denoising strength. The famous 10% finalization that takes 90% of the work time (for a human). This is only the beginning..!

[horribleCodes]

It will completely transform product and architecture design. Make a rough sketch or blocked out model in SketchUp, and immediately generate details, shading, backgrounds, props,...
Also, generating tileable textures for CGI.

[cmp-nct]

Out of the true simple joy of being able to generate images from a few well chosen words, rivaling with the best artists in the world, I think a very important subject is "what is this really useful for ?"

• Video games / CGI

  • "generating portraits for a game mod of Age of Empires 3 Definitive edition" (TY @matrix4767)
  • tileable textures (TY @horribleCodes)
  • sketch2img (TY @horribleCodes)
  • Tile finalization (TY @hydropix)

• Security

  • Anonymization of faces / persons / scene in photographs for social networks and other image-based communication

I'll begin, then expand, this list everytime I find real use cases for this.

I believe the real question goes beyond "what is this useful for?" and more towards "What is this going to be used for in the future?".
I am not a philosopher, but I believe that the implications of generative art are far reaching and will have a profound impact on the future of humanity.

We've had transformative technologies in the past, such as the printing press, the steam engine, the internet, the transistor, the computer, the smartphone and all of these have had a profound impact on the way we live our lives.
Generative art is the next transformative technology and alongside it's stunningly forceful introduction several other machine learning technologies will be introduced.
Think about generative music, generative writing, generative movies, generative medicine ( a computer will diagnose) and generative law ( a computer will write, explain and interpret the law).

The current models are so flawed you need to become an expert to generate proper images with it, or very patient.

Right now we think about tileable textures .. but do you need to tile and repeat a texture if you can dream one up seamlessly and unique ?
Artists are already going extinct, and the next generation of artists will be replaced by machines with the exception of speciality artists and very high end ones.

Think about photography, who needs a beautiful model when you can generate one with a few words or transfer one a posing random person ?
The same for logo/icon designers, simple forms of architecture, illustrators, caricaturists, manga/anime makers and countless more visual talent infused jobs. It's all going down the drain.

I expected things like this to happen for more than a decade but I did not expect it to happen with such a bang. The entire world is being transformed. It's going to happen fast and broadly.

Also 60% of this text was not written by me but by a generative model. I just guided it.
You can spot my own writing by the fact that it's not very good :)

[hydropix]

And I translated it with an AI... (as my answer)^^
I don't believe in artist obsolescence at all, but some manual skills are no longer needed and, in fact, artists who have only manual skills are screwed. This is not the case with artistic direction.

[cmp-nct]

@hydropix A few cues what led me to agreeing with the AI-philosopher's opinion :)

1. Imagine you are a talented illustrator creating sketches for newspapers, magazines, book covers.
   You likely put your work up on a few websites so people can choose them and if they love it and want a unique one they contact you.
   You are now competing with your work, each one takes days to finalize, against an AI that finalizes in a fraction of a second. You are overwhelmed by competitors now.
2. Like you said "some skills" no longer needed. Those skills are the foundation of artists. It's the talent to imagine something, realize your thought on paper/computer in a way that other people find pleasing or disturbing. That very rare talent is what made an artist successful.
   The only skill you'll soon need with AI art is the "receiving end". You need a taste for "pleasing or disturbing" content, you don't need the creation skill anymore.

This is what destroys the vast majority of artists. They are gone already, you just see their shadows now.

There is still room for some of them. The outstandingly gifted ones, the famous ones and those who create art that can not (yet) be done by AI. For example a photographer of marriages or in clubs or of products. Anything that has a "local" connection.

It certainly has a dark side, that's why many people oppose the idea that artists are gone forever.
Artists have been a foundation of human civilization since we can speak.
That side is there. However there is an upside too, it's empowerment of people.

My personal weakness in IT is art and design. I know how to use photoshop, Maya and other tools but I lack in the creation process. I have a well rooted taste for visual art but I have no talent in creating something like that.
The AIs turn the table, suddenly my taste is all I need. So I am personally gaining abilities I didn't have before, the minority (artists) lose that advantage.

[hydropix]

I agree with all of this. But with some nuances, I'm an artist myself, for several decades (https://www.artstation.com/hydropix). And I have many professional friends who are now using AI with talent, in the video game or movie industry. I notice that for the moment, they use it much better than non-artists. They have the eye, and the culture to guide the AI. And also we are able to make the fine adjustments on the final image. Retouching with the same level of quality as the AI is not accessible to everyone. Of course in the future this step will probably not be necessary anymore. What I want to emphasize is that we should not believe that everyone will be able to manage AIs with talent. The quality of images will increase enormously, but in the global competition, the top will remain the top... (a new Top) with new skills and some old ones that are difficult to access. It's all about competition all the time. The best will always be rare.

[cmp-nct]

Nice paintings :)
I agree that today there is talent/skill required to produce good results, from prompt engineering to polishing.
Right now it requires a lot of knowhow and understanding to produce good AI results.
But my comment was looking slightly into the future, these borders are going to fall quickly. Results will be flawless and much better controllable by normal language. Hard to say where it will end but I imagine an interactive process, possibly by voice.
So you speak to the computer and it dreams the image in realtime while you talk, the hardware to do that so quickly might take a couple years. Looking at Copilot it's probably already doable in realtime today if you run it on the right cloud infrastructure.

[Ehplodor]

"Victorian house in a dream digital art by hydropix trending on artstation" 😇👏😍🎨🖌️

[Chilluminati91]

Im a music producer and share my work on youtube and sell licenses on various platforms. I used to take a picture of a rapper with a similar sound (type beat) and put it in the video. That is not exactly legal. Now I just generate a dope image of said person with Stable Diffusion. Looks way better, gives me creative freedom to deliver my artistic vision behind the song and doesnt put me in legal trouble.

So AI helps me with marketing until Music-AI is developed enough to take away my job.

[Ehplodor]

I heard this morning on the radio about a legal case about a photo of Prince, the artist, that was sold for about 400 $, then modified by Andy Warhol. And now a dozen previously unpublished similar pieces of art are in the center of a battle about whether modified content is still copyrighted or not... Or something like that. This seemed so much relevant to the img2img thing and variations and alternatives... A few months to wait for the final verdict.

[cmp-nct]

Though that legal case is not going to mean anything to us, it's a specific case.
The only legal case that would change something is when a high court decides that there is no copyright on AI art and that's not possible.

Copyright is always case by case.
Regardless if you use img2img or txt2img, in both cases you can violate copyright.

If your output resembles copyright protected art or trademark protected stuff then you are violating it, even in txt2img.
If your output is similar to an original composition of copyrighted work, then you might violate it even if you made something totally different.
But for every single image that's an own case and a court has to decide if the borders for copyright are broken or not.
It's not a problem of AI art, it's a general problem.

[iioaia]

Dematerialize content generation.

[Ehplodor]

Hi, thank you for contributing ! Would you mind elaborate a little on this subject ? I'm unsure about what you mean actually.

[ClashSAN]

-Makes weird pictures as a flashcard memorization tool.

[Ehplodor]

Making puposefully weird thought provocating pictures for better memorization XD I love this.

[Doev]

Hard to say.

In my opinion it is a powerful tool that allows people with more technical skills to create artistically high-quality works.

So far it's just a toy for me. But I probably haven't been able to look at the many freely available ideas yet.

I think there's a lot of potential there. The problem I see at the moment is that it's difficult to create a large amount of content that is internally consistent.

[18292722059]

你好，杨宝强已经收到了您的来信。 Hello! baoqiang have received your letter.

[miaomi8]

这是来自QQ邮箱的假期自动回复邮件。   您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。

[bridgebrain]

I've been using it to do concept sketch with clients (video, website, logo design), and after we work on it together for a bit, I hand them a huggingface space (usually diffuse the rest or 2 turbo) and have them spend a while exploring to try and get what they want out of it. I find it helps them hone in on the things they're actually requesting from me, and they can hand me some good examples. It's becoming a lifesaver with the "I'll know it when I see it" crowd and lets the "5 million changes" get their thoughts out before coming to me with them