Using AI for cover art

[TM2YC]

Then a quick and dirty photoshop with the most promising one gives me this:

https://imgur.com/y8DEMgB

^ I think this is a good way to go. It looks like your original stark idea but with an extra bit of smooth style for free. Using AI as a brainstorm session and then make something yourself inspired by the ideas it flings out there.

 

[TM2YC]

I was curious to have a go with the "combine AI stuff with my own judgment" type idea. Since I just added a Lord of the Rings edit to ifdb, I went with that franchise (this poster took me less than 30mins to make from start to finish).

1. I brought up the first free AI art thing suggested by Google: https://hotpot.ai/

2. After a bit of trial and error with word prompts, "scary fantasy landscape mountain volcano tower ash wasteland" generated something I thought looked like Mount Doom:

3. Since it was only 500x500 and I couldn't be bothered to sign in, or pay. I upscaled that to 2K using another free site: https://www.iloveimg.com/

4. I did a Google image search for "sam and frodo mountains" and "hobbits silhouette" stuff like that and came up with:

5. Google image searched for Return of the king title and found one on Wikipedia:

6. Put it all together. Desaturated sky. Adjusted contrast/brightness in a few places. Took Sam, Frodo, Gollum and the Nazgul from that image and dropped them into an appropriate area where the light draws the eye. Few more tweaks, FE.org log and here it is:

 

[tremault]

a lot of these images have elements that experienced artists will be able to pick out right away. I think this can be problematic if people who are against AI see your cover and just turn their back on your work based on that. I used AI for the first version of my Kang cover, as I couldn't find stock images for what I wanted, so I generated an antman figure and a small army. I regretted it in time because I knew I would lose some potential viewers, so I spent some time and found images that suited the purpose. I don't think AI images are terrible as placeholders, but if you can, I strongly encourage getting a human to replace the AI elements. Art is communication and AI doesn't really communicate anything. It's the human element that we need.
That LOTR image for example, I'm glad the floating mountains were cut out, although they are still partly visible. I think the image you presented communicates, thanks to your layout, but the whole AI makes me uneasy from the off.

 

[MightyAttackTribble]

That is an excellent bit of work.

Wherever one may stand on the morals and ethics of using AI art, let's not lose sight of the fact that we are a community based around a hobby predicated on making unauthorised edits of other people's works. We don't do it for profit – indeed, I suspect most of us have spent decent sums of money to pursue this hobby – so I don't see any harm in using AI art to create covers for our projects.

 

[tremault]

That is an excellent bit of work.

Wherever one may stand on the morals and ethics of using AI art, let's not lose sight of the fact that we are a community based around a hobby predicated on making unauthorised edits of other people's works. We don't do it for profit – indeed, I suspect most of us have spent decent sums of money to pursue this hobby – so I don't see any harm in using AI art to create covers for our projects.

I'm not sure I understand how you're using this to support your point. I don't see how that is a defence, in fact it highlights the fact we are non profit creators and the primary desire for creatives who aren't in it for money is attribution. When we edit using a movie and other sources, our priority is to ensure we respect the rights of the creators. AI cannot protect the rights of the original works it used to craft it's model. attribution is impossible. The produced image is a homunculus of all works in the corpus that included those key words. There's an artist who painted a mountain, whose soul is embedded in that image of the volcano, but it's so far removed that their skill and life experience is mixed into the soup as it were. it's completely obfuscated.

 

[MightyAttackTribble]

When we edit using a movie and other sources, our priority is to ensure we respect the rights of the creators.

And yet, we do not.

The creators have not given us leave to use their work to create derivative works.

They have not given us permission to distribute those derivative works – quite the opposite, in fact – and this site has to tread extremely softly to avoid getting shut down by those rights holders.

We steal art and make it our own, after a fashion. How is adding one more piece that in most cases will rarely be seen outside of this community any worse?

Generative AI has issues that need addressing, but whichever way you look at it we do not occupy any kind of moral high ground.

 

[tremault]

And yet, we do not.

The creators have not given us leave to use their work to create derivative works.

They have not given us permission to distribute those derivative works – quite the opposite, in fact – and this site has to tread extremely softly to avoid getting shut down by those rights holders.

We steal art and make it our own, after a fashion. How is adding one more piece that in most cases will rarely be seen outside of this community any worse?

Generative AI has issues that need addressing, but whichever way you look at it we do not occupy any kind of moral high ground.

I don't think i was suggesting we occupy any high ground, but my point was that at the very least, we give credit to the original creators of the works we cannibalise. With AI art, we can't even do that. This could be an extension of the phrase "Honour among thieves".

edit: actually I don't think it's fair to use the word steal. We take our own copies that we own and we transform them under fair use and we share those copies with people who also own their own copies.

 

[MightyAttackTribble]

I don't think i was suggesting we occupy any high ground, but my point was that at the very least, we give credit to the original creators of the works we cannibalise. With AI art, we can't even do that. This could be an extension of the phrase "Honour among thieves".

In an ideal world, yes, we would be able to credit everybody. At the very least make it known that generative AI was involved and that I am not claiming authorship.

Still, in terms of generative AI in this particular application, I have my doubts that the original artists would care one way or another. I have creative works out in the world, nothing particularly impressive but I do hold the copyrights for them. I'd be greatly pissed if I found someone to be profiting from them financially in some way, but I'd not give any thought to somebody using them for something that very obviously isn't.

I know I cannot speak for everybody in that position, just as any one person involved in a movie I edit cannot speak for the thousands of others who worked on it cannot, but if done from an honest position of "This is a derivative work, I claim no ownership nor authorship, and will attribute or remove any works as requested" I don't see the harm.

 

[unfair]

Personally if I'm going to be slapping my name on some art, I'm a lot more comfortable with it being some frankenstein procedurally generated work than a custom fanart that someone poured sweat and tears into and I just slapped a title on because I thought it looked cool. AI art is a complex issue, but unless you're actually reaching out to these creators and asking their permission to use their art as a cover for your edit, I see that as far more invasive and potentially upsetting to an individual than the vague accusations about AI training data by people frightened that AI will replace their work.

In one situation you're taking wholesale something that someone definitely owns, and in the other you're fretting about the rights of an image that has unspecified ownership ...and placing that worry as greater than the act of essentially commercializing (in a non-monetary sense) a piece of fine art without permission and saying it's okay because you gave credit.

If someone comes to me and is able to show that the AI stole an identical element from their art and used it in my cover then I'm happy to take it down if it upsets them, but that scenario seems highly unlikely for quite a few reasons, so this is a purely theoretical moral objection and not some legal or human impact that we're discussing, and I think that sort of high level decision about AI output should be left to the court.

 

[tremault]

Still, in terms of generative AI in this particular application, I have my doubts that the original artists would care one way or another. I have creative works out in the world, nothing particularly impressive but I do hold the copyrights for them. I'd be greatly pissed if I found someone to be profiting from them financially in some way, but I'd not give any thought to somebody using them for something that very obviously isn't.

I hadn't really considered that here yet, but I do have a not insubstantial amount of work on the net, a lot of it is fanart for a franchise I love and my work was used in a youtube video with zero attribution. I object to that and I asked the video maker to attribute my work. Sure, my work is built on someone else's, but that's true of literally everything. There's nothing original in the world.

I also object to my work being used to train ai. When I learnt that Deviant Art was going to be using people's galleries to train ai, I deleted my account and removed all my work from their service.

 

[tremault]

If someone comes to me and is able to show that the AI stole an identical element from their art and used it in my cover then I'm happy to take it down if it upsets them, but that scenario seems highly unlikely for quite a few reasons, so this is a purely theoretical moral objection and not some legal or human impact that we're discussing, and I think that sort of high level decision about AI output should be left to the court.

I am pretty sure it would be absolutely possible to recreate an artist's original work if the random seed was removed from the input and if the exact description assigned to that original image was fed in.

also, I would much rather somebody use my work and give me the credit for what is mine, than use a portion of my work fed through a filter and not credit me at all. people have used my art on a few occasions and they credited me and I am happy about that because it potentially brings their audience to me and the rest of my work. I make art to share it with the world, not for an algorithm to dissect and disseminate in pieces.

 

[unfair]

I am pretty sure it would be absolutely possible to recreate an artist's original work if the random seed was removed from the input and if the exact description assigned to that original image was fed in.

That was my point, this is a purely theoretical argument because your proof of harm is to remove everything that is core to a generative AI and just tell it to recreate the training data. So while I understand that some people have an intellectual objection to AI, and that's perfectly fine, I don't really think this is the venue to hash out those feelings since they don't have concrete relevance to cover art.

 

[addiesin]

I go back to what I said in the other thread where this came up. Using AI for a fan edit cover and saying, "I used AI for this", is better than stealing real fan art and saying, "I made this"or saying nothing about where it came from, which has been the standard practice.

 

[tremault]

That was my point, this is a purely theoretical argument because your proof of harm is to remove everything that is core to a generative AI and just tell it to recreate the training data. So while I understand that some people have an intellectual objection to AI, and that's perfectly fine, I don't really think this is the venue to hash out those feelings since they don't have concrete relevance to cover art.

You're right. I made the mistake of assuming this was invited, but in actuality I'm derailing your topic. I'm not sure how I ended up in that head space, sorry about that.

 

[unfair]

You're right. I made the mistake of assuming this was invited, but in actuality I'm derailing your topic. I'm not sure how I ended up in that head space, sorry about that.

It's all cool! I don't mind little interesting rabbit trails, just seemed like all the opinions on both sides had been expressed, and without a concrete answer it's just a topic for pondering and keeping an eye on until we see the evolution of the legal and regulatory system surrounding it.

 

[TM2YC]

That LOTR image for example, I'm glad the floating mountains were cut out, although they are still partly visible. I think the image you presented communicates, thanks to your layout, but the whole AI makes me uneasy from the off.

Yeah, I was more interested in what could be produced in under 30 minutes, it'd take me more like 30 hours to actually get out my art materials and paint Mount Doom myself (I have hand drawn some of my fanedit art before). I thought about removing those mountains in the back, the birds in the sky, the waterfalls and lake, the pyramid of Giza that weirdly appears to be at the base of mt doom and adding an evil looking silhouette of a tower to the right mountain (Like Cirith Ungol) and/or Baradur into the distance but again I was interested in exploring the speed and ease of using AI for fanedit art.

I doubt I'll use AI myself for fanedit posters because I enjoy using my own creativity/skills but this is clearly a viable option for people who can't, or who think this is more ethical that simply lifting other people's work without their permission wholesale (as has been the standard before). I think it'll be most useful in the way I used it, by combining copyrighted franchise elements like Frodo/Sam etc and a generic mountain generated by AI. Or "Batman looking over AI city scape", or "Millennium Falcon flying through AI galaxy", or "Kirk and Spock exploring AI alien planet" etc.

 

[WilliamRedRobin]

Thanks for the advice @unfair
I managed to make this poster for a matrix edit using you method:

Poster hosted at ImgBB

Image Poster hosted in ImgBB

ibb.co

A little more time in photoshop could probably make it a lot better, but it's already a big improvement on what I was doing before.

 

[unfair]

Attempting a cover for my The Batman edit... apparently it is possible to trick the AI into providing certain recognizable figures if you phrase it right, as this test shows. I generated several dozen images, almost all of them were on point but I'm only going to include the best or most interesting.

First prompt "superhero bat on skyscraper rooftop overlooking city":

https://imgur.com/DlrXn3J

https://imgur.com/UhC5pgl

Not exactly what I was looking for, looks like it's combining superman and batman in some of the images. Let's try "night brooding superhero bat on skyscraper rooftop looking down over city":

https://imgur.com/QZLyJ7X

That's a little better, but positioning is a bit janky and characters look better from side or behind so let's go with... "night brooding superhero bat from behind on skyscraper rooftop looking down over city":

https://imgur.com/PdFq8Ah

https://imgur.com/bDeeFGH

https://imgur.com/CRdA9TK

https://imgur.com/3deNhc1

https://imgur.com/vqdkKpZ

https://imgur.com/sZoeEyz

And the final cover:

https://imgur.com/dtIteGh

Edit: Tweaked font outline color on final image a bit

 

[WilliamRedRobin]

Couldn't resist this.

Prompt: "Silhouette of a bat-themed superhero merged with the skyline of a city in chaos"

+ Pixlr's generative expansion feature and some custom fonts:

​

I better stop now before I make posters for every edit on my ideas thread

 

[WilliamRedRobin]

Ok, one more:

​

This is definitely my favourite way of making posters now. It feels a lot better than just taking someone else's poster and putting a title on it, and it looks a lot better than when I try to make my own posters.
It's not what I'd use for every edit I make, but there's definitely ideas for edits I have (like these ones) that it suits perfectly.