Generative AI FAQ (MidJourney, Dalle2, SD) đ¤
There are certain questions that come up over and over again with AI:
đ What are the tools you use the most?
đ Copyright. Who owns the images?
đ Can I legally imitate styles?
Etc.
Here you have the main QUESTIONS & ANSWERS đ
There are certain questions that come up over and over again with AI:
đ What are the tools you use the most?
đ Copyright. Who owns the images?
đ Can I legally imitate styles?
Etc.
Here you have the main QUESTIONS & ANSWERS đ
1. What are the tools you use the most? Where do I begin?
Here's a list. But beware, this changes VERY quickly. It goes at the speed of light:
Here's a list. But beware, this changes VERY quickly. It goes at the speed of light:
2.1. Copyright. Who owns the images?
There is no jurisprudence in this regard, so unless the terms of use of the platform where you generate them say otherwise, in general terms, the copyright on the images belongs to whoever generates them.
Continue đ
There is no jurisprudence in this regard, so unless the terms of use of the platform where you generate them say otherwise, in general terms, the copyright on the images belongs to whoever generates them.
Continue đ
2.2. The main platforms such as StableDiffusion, MidJourney and Dalle-2 make it clear in their terms of use that they assign the authorship of the images to the users who generate them.
2.3. If you live in the US: there is supposedly (not entirely clear) no copyright protection for solely machine-generated works. BUT, copyright is possible in cases where the creator can prove that there was substantial human involvement.
2.4. For example, in the USA, issues related to a work (a comic, "Zarya of the Dawn") that has been recently rejected by the copyright office have come up several times in the press.
2.5. The artist has filed an appeal, since she alleges that the direction of each cartoon of the comic, as well as the common thread of the story she invented, proves "substantial human participation".
We'll see how it ends!
We'll see how it ends!
2.6. If you do not live in the USA, today, none of this applies to you.
And if you do, or if you still have doubts, just change the generated image enough so that the authorship is clear that it's yours.
And if you do, or if you still have doubts, just change the generated image enough so that the authorship is clear that it's yours.
2.7. And in the worst case, imagining a world in which other people can take advantage of the images/videos/creations that you have generated for your projects using AI in their own creations, does not seem like a drama to me personally. Open Source taken to the extreme.
3. Does this mean I can create an AI Super Mario comic and sell it?
Of course not!
Characters like Mickey Mouse, Spiderman, Mario, etc. They are all protected by IP. It doesn't matter if you draw them in charcoal or with AI... you can't sell a work in which they appear.
Of course not!
Characters like Mickey Mouse, Spiderman, Mario, etc. They are all protected by IP. It doesn't matter if you draw them in charcoal or with AI... you can't sell a work in which they appear.
4.1. Can I at least imitate styles? I would like to make a "Studio Ghibli" style comic.
Yes you can!
Styles are not copyrighted. In fact Studio Ponoc, made up of ex-Studio Ghibli employees, is making Ghibli-style movies without any problem.
Yes you can!
Styles are not copyrighted. In fact Studio Ponoc, made up of ex-Studio Ghibli employees, is making Ghibli-style movies without any problem.
4.2. Another example: do you remember the movie Anastasia? Sounds like Disney right? Well, no. It was from Fox. And there is no problem
Fortunately, the styles cannot be copyrighted, or the design and art would be in perpetual lockdown.
Fortunately, the styles cannot be copyrighted, or the design and art would be in perpetual lockdown.
5.1. Are the AIs copying? Do they make "collages"?
I leave you with this video so that you have an explanation of how they work internally, specifically, the concept of latent space.
youtu.be
I leave you with this video so that you have an explanation of how they work internally, specifically, the concept of latent space.
youtu.be
5.2. As the video explains, AIs do not copy or collage.
They "observe" a large number of images and "learn" to break them down into high-level patterns that make them up (lines, composition, etc).
Then, based on these patterns, they can merge them to generate new ideas.
They "observe" a large number of images and "learn" to break them down into high-level patterns that make them up (lines, composition, etc).
Then, based on these patterns, they can merge them to generate new ideas.
5.3. In the future, its number of inputs could be even greater: three-dimensional models of the human body and of any animal to perfectly understand their anatomy, "eyes" in the real world through cameras, sounds, music, etc...
6.1. Why do they sometimes generate signatures/watermarks?
Because they understand that it is a high level pattern.
If for example the AI ââhas observed drawings from 1800 with signatures, it may tend to try to generate imaginary signatures.
Because they understand that it is a high level pattern.
If for example the AI ââhas observed drawings from 1800 with signatures, it may tend to try to generate imaginary signatures.
6.2. Where does it get the signature from?
The AI invents it!
In the same way that it generates a new face, the AI ââsometimes generates signatures that we could be considered as âan average of some of the observed signaturesâ.
The same with titles, logos, pineapples, shoes...
The AI invents it!
In the same way that it generates a new face, the AI ââsometimes generates signatures that we could be considered as âan average of some of the observed signaturesâ.
The same with titles, logos, pineapples, shoes...
7.1. Is it possible for an AI to generate two identical images?
No, diffusion models, even using a seed number (with which you can force the image to render almost the same) are NON-deterministic, so there will always be some small differences.
No, diffusion models, even using a seed number (with which you can force the image to render almost the same) are NON-deterministic, so there will always be some small differences.
7.2. I'm not sure at which layer (software? hardware?) it arises this random effect. I haven't gotten anyone to answer that question yet.
8.1. What limitations do they have today?
Something already famous is the problem they have with hands: they are horrible đ. They are also not good with texts/fonts.
And in general, they have a lot of small flaws that many like to point out.
Something already famous is the problem they have with hands: they are horrible đ. They are also not good with texts/fonts.
And in general, they have a lot of small flaws that many like to point out.
8.2 But keep in mind that just a year ago they were only able to create an icon of an avocado chair... and today they can generate a still from an imaginary movie, with invented actors.
Maybe by the time you read this FAQ, we'll be at the point where they're indistinguishable.
Maybe by the time you read this FAQ, we'll be at the point where they're indistinguishable.
8.3. It is only a matter of time before their generations will be perfect and indistinguishable from any human-created illustration, rendering, or photograph. And after that, video and real time.
9. Are they art? Are they creative?
Ugh, those debates are sooooooooo 2022 đ
"Art has the lovely habit of ruining all artistic theories." â Marcel Duchamp.
Actually, it doesn't matter one bit. They exist.
Ugh, those debates are sooooooooo 2022 đ
"Art has the lovely habit of ruining all artistic theories." â Marcel Duchamp.
Actually, it doesn't matter one bit. They exist.
10.1. Where is all this moving to?
I shouldn't put predictions in a FAQ, but... hey, it's my FAQ đ
I think that the industry of design, entertainment, video games, cinema, etc. they are going to experience a transforming tsunami in the coming years.
I shouldn't put predictions in a FAQ, but... hey, it's my FAQ đ
I think that the industry of design, entertainment, video games, cinema, etc. they are going to experience a transforming tsunami in the coming years.
10.2. I believe that all design suites will end up integrating AI and that soon working with AI will not be simplified to write a "prompt" and wait for a generation. We will be able, IN REAL TIME, to control all the objects, characters, details, style, lighting, materials, etc.
10.3. In fact, though rudimentary, plugins are already coming out like hot cakes. And the way of working with them is not "I prompt and that's it". No, it is a complete work process.
But in the future, it could be be more similar to a 3d program!
youtube.com
But in the future, it could be be more similar to a 3d program!
youtube.com
10.4. And by being able to run them in real time, not only will come a new way of designing, but also we will see video games and CGI whose visuals will not have a polygonal base, but will be generated pixel by pixel in real time. Can you imagine? Extreme generative video games!
10.5. Even a 5-year-old child, talking and moving his little hands, will be able to create fantasy worlds.
That is where we are moving to!
By the way, I'm sure this FAQ is going to get out of date in almost all its points very quickly.
This is going very fast!
That is where we are moving to!
By the way, I'm sure this FAQ is going to get out of date in almost all its points very quickly.
This is going very fast!
11. Is there a guide to get started with generative AI?
I'm glad you asked! đ¤Ŗ
I'm working on a prompting guide that you will love. You can subscribe to my newsletter to know about it as soon as I will publish it! Hopefully next week!
javilopen.substack.com
I'm glad you asked! đ¤Ŗ
I'm working on a prompting guide that you will love. You can subscribe to my newsletter to know about it as soon as I will publish it! Hopefully next week!
javilopen.substack.com
Tim added this clarification about the comic "Zarya of the Dawn".
I think she is right. I believe there is a lot of "substantial human participation" in her work.
I think she is right. I believe there is a lot of "substantial human participation" in her work.
What are your thoughts about this tech and the changes society is already experimenting?
I would love to read your opinion đđ
I would love to read your opinion đđ
Loading suggestions...