Jump to content

Adventures in A.I. - discussion and your creations


PeteJ
 Share

Recommended Posts

i-MnQFXC5-X2.png

Adventure through an AI Mind. Conceptual. Fantastic. Aspect Ratio 21:9

 

In off topic we have a thread dedicated to the meme generator that is DALL.E Mini, head over there for some comedy examples.

 

I thought I'd create this thread as more a creative outlet, as well as a discussion point around AI generated content. If this proves popular I'll put up some guides as well.

 

In my opinion some of the output from AI is getting staggeringly good, which on one hand is fascinating but on the other is concerning. If it becomes so easy to generate 'art', what becomes of the value of art? Who owns the copyright to generated images?

 

Here are some examples I've come up with, which are 100% AI generated. The text under the image is what I asked for.

 

Image generation by MidJourney - https://www.midjourney.org/

Image up res by Topaz Labs Gigapixel AI - https://www.topazlabs.com/

 

If anybody is interested I have a handful of trial invites to Mid Journey.... if I can figure out how to use them. You'll need a Discord account.

 

i-2h45zXJ-X3.png

Adventure through a magical forest. Painted. Colourful. Adventurer

 

i-s9SqkPk-X3.png

Bluebell woods. Classic painting. Path winds through scene. Colourful.

 

i-fjBv8Q3-X3.png

Rain covered futuristic cityscape. Concept art. Reflections. Lights. Crowded. Skyscrapers. Flying cars

Link to comment
Share on other sites

One of the things I like is how one AI art prompt can be interpreted in many different art styles. For example, these three images were all created by the prompt ‘red haired girl with squid fingers’:

 

207042A5-4770-45E1-B5B7-C81F2AB1BC65.thumb.jpeg.9c5997e4ec6d1a7b72d6cb5f08ffaf77.jpeg00A68A12-E0C8-4369-A4E5-026C6541E179.thumb.jpeg.3115274ae9cc315d38569d88882ab27a.jpeg45C70853-5458-4F46-A364-B324495F1CCA.thumb.jpeg.e74eb8da9d81ff9cb18cfc15e268e869.jpeg

Link to comment
Share on other sites

On 01/07/2022 at 14:07, PeteJ said:

 

In my opinion some of the output from AI is getting staggeringly good, which on one hand is fascinating but on the other is concerning. If it becomes so easy to generate 'art', what becomes of the value of art?

 

I do find these quite worrying, because clearly they've created artificial intelligence, in so far as it's taking data and information it's been taught, and applying context to it to create a response to a linguistic prompt, but I always thought that art was so much more than that, and required something 'special' that you only find in humans. Sentience, consciousness, emotions, a soul, whatever. And it's looking like maybe it isn't?  Even some of the more esoteric DALL-E Mini ones - things like "the aesthetics of anger" rather than "boris johnson eating a sausage roll lol" - if you didn't know they were AI-generated you (or at least I) could look at them and think "Oh, that's quite an interesting abstract take on that concept by this artist".

 

Could you conceivably teach an AI like these all about every genre of music and the intertwining styles and influences, and then type in "Chicago House tune with a jazz sample" and have it spit out something with just as much musical merit as if a human had made it?

 

It's worrying! I'm worried!

Link to comment
Share on other sites

I think all the ones I've seen reference existing artists. So it's not bleeding edge, it's like illustration commissions where you're given mood boards and artists they'd like you the reference. It's not like going to a really fucking good art show where it's all surprising and unfamiliar.

 

But also I have a question. I need an in for generating cutaways in my next film. I need that creepy style of dog dream google deep mind stuff, but with a slow dolly over the surface of a road in the british countryside. I think I need a machine learning approach so I can breif it on what images to reference. So I want hang-on and outrun, and live action footage of british lanes at the right camera position. Where do I start for that?

 

Link to comment
Share on other sites

On 01/07/2022 at 22:07, daninski said:

But also I have a question. I need an in for generating cutaways in my next film. I need that creepy style of dog dream google deep mind stuff, but with a slow dolly over the surface of a road in the british countryside. I think I need a machine learning approach so I can breif it on what images to reference. So I want hang-on and outrun, and live action footage of british lanes at the right camera position. Where do I start for that?


AI video is still in its infancy, if you want a glassy eyed bloke to badly lip sync for you then AI has you covered. AI is used in intermediate steps, like if I need all the people in a scene subtracted for a camera track then AI is amazing for that or if you want to change one face for another obviously but for what you’re suggesting - essentially a video version of MidJourney I’m not sure it’s really possible just yet. 
 

One thing you could try is to create a high res still using AI and then reproject that onto a plane with a moving camera? 

Link to comment
Share on other sites

10 hours ago, petrolgirls said:


AI video is still in its infancy, if you want a glassy eyed bloke to badly lip sync for you then AI has you covered. AI is used in intermediate steps, like if I need all the people in a scene subtracted for a camera track then AI is amazing for that or if you want to change one face for another obviously but for what you’re suggesting - essentially a video version of MidJourney I’m not sure it’s really possible just yet. 
 

One thing you could try is to create a high res still using AI and then reproject that onto a plane with a moving camera? 

The creepy unsettling thing I’m after mostly is that thing where you start to see things without seeing them appear. The deep mind dreaming of dogs footage for example. I’m looking for something swirling and eddying like that. And also at a high frame rate to really make the room queezy. 

Link to comment
Share on other sites

Here's some more of my latest creations...by which I mean entering some text into a field and pressing enter.

 

i-PP6cSDh-X2.png

coloured paint being splattered over a black and white picture of a highland cow. subtle paint running down image. colour splattered everywhere. colourful. bright. artful. aspect ratio 16:9

 

i-rZ25X64-XL.png

adventures in the metaverse in the style of fantastic voyage. colourful.

 

i-9vPfCfN-XL.png

adventures in happiness. adventurer. aspect ratio 16:9

 

i-mLH9z4J-XL.png

procrastination

 

i-gjPZRtC-X2.png

Adventures through time and space. concept painting. creative. freedom. bursting with colour. aspect ratio 16:9

 

Link to comment
Share on other sites

I tried with the same prompt as you from above ('Adventures through time and space. concept painting. creative. freedom. bursting with colour. aspect ratio 16:9'). First question, how are you making what you use respect the aspect ratio? I'm having to go through the Discord bot for the Midwinter beta, and it never comes up with anything but 4:3 even if you request a specific aspect ratio.

 

Midwinter:

Michael_R_Grant_Adventures_through_time_and_space._concept_pain_244de1bd-5e64-40a2-b481-7597a31f72ac.png.4b651e2f0f81dad36b1415585c5e9d43.png

 

Big boy DALL-E:

 

1696176233_DALLE2022-07-0517_12.11-Adventuresthroughtimeandspace.conceptpainting.creative.freedom.burstingwithcolour.aspectratio16_9.thumb.png.496205dd9ecdbf1a82d1b7a6f290b5fd.png

 

For a laugh, here's DALL-E Mini: 

 

download.png.422644d3546988765d3d8bdb12ec0362.png

 

Midwinter is clearly the best here.

Link to comment
Share on other sites

Yeah mid journey is very cool for art, the discord hall of fame channel has some great stuff in it. My text for aspect ratio isn't entirely true, since I didn't think people seeing the text would understand it. 

 

For aspect ratio use 

--ar 16:9

(Or any aspect ratio)

 

You can also do things like

--uplight

To reduce the impact of how much the upscale command changes the original image. You can often find the upscaled version adds additional, unwanted, detail

 

Or

--no blue

To not have any blur in the image. 

 

NEW

--test 

Uses the new routine which gives better results, especially in portraits 

 

::Yellow

Add weight to any part of your prompt by prefixing the word with ::

 

At the top of discord there is a pinned message button, which has a link to the guide with the parameters in it. 

Link to comment
Share on other sites

Guess who didn't realise there were a limited number of images available via the free trial... :lol:

 

I now have a lot of images using the following string:

 

Lovecraftian, temple, dank, subterranian, frightening, creatures, --ar 16:9

 

90a6c790-828e-409f-bdff-0a5a3c4cc728_FishyFish_Lovecraftian_templedanksubterranianfrighteningcreatures.thumb.png.7150cf243b43d766ddbda174af3d1c25.pngFishyFish_Lovecraftian_temple_dank_subterranian_frightening_cre_184dab4f-1b1a-4cc4-9d06-2914d2975036.thumb.png.c32a010279ff4378ca5537b9ace539ee.png13b1c7d2-349c-40d5-97e6-7ab2bf79deda_FishyFish_Lovecraftian_templedanksubterranianfrighteningcreatures.thumb.png.5e803e4e1bebab601d857ede9ec5a094.png89044e68-4e87-4e9f-9e37-e7f4416ebe65_FishyFish_Lovecraftian_templedanksubterranianfrighteningcreatures.thumb.png.190423e192f7839dd2d5e8ffabea3948.png

Link to comment
Share on other sites

It's very, very addictive. I've nearly burned through a whole month's worth of credits in less than a week :(

 

i-Hnw8px7-X2.png

cyberpunk city at night. abstracted painting. primary colour yellow and black. People looking over city. aspect ratio 16:9

 

i-p5LW337-X2.png

a picture of a dragon in the style of monet

 

i-GCT29zc-X2.png

different shades of blue in battle

 

i-zF28JD9-X2.png

crowd celebrates under neon rain

 

i-FqxSjLz-X2.jpg

samurai walks through a bamboo forest

 

 

Link to comment
Share on other sites

On 01/07/2022 at 16:59, PK said:

I always thought that art was so much more than that, and required something 'special' that you only find in humans. Sentience, consciousness, emotions, a soul, whatever. And it's looking like maybe it isn't?

 

This is very interesting (and as has been said, concerning). Especially because as I browse this thread I have found myself feeling like I sense emotion in the art. I have been worrying for years that something like my musical tastes (for example) would actually probably be very easy to recreate without any kind of human interference if a machine just studied all the songs I listen to and mimicked the style very well. But then, of course, as with the pictures in the thread, they wouldn't have been possible without the initial pieces and people that inspire them. But then again couldn't you say the same for all human art as well?!

 

I guess maybe art is two things. Maybe even distinct. One, a unique act of emotional creation that could only have been specifically achieved by a unique person or group of people. And then once it's finished it's all about the interpretation of that creation. Everyone who experiences it perceives it in a slightly different way and it becomes many, many different things in different people's memories/consciousness.

 

I studied AI at uni ages ago and these kind of philosophical angles were my favourite part of the course to be honest. It was only a couple of modules but studying machines to learn more about what makes us human seemed fascinating to me.

Link to comment
Share on other sites

A simple prompt of just 'infinite possibilities'. Multiple re-spins to keep honing on on the 'right' image.

 

i-DzvkMMW-XL.png

 

Attempting a realistic photograph of a person. Not bad, still clearly computer generated though. Ears are especially off.

 

i-pM4KK7B-XL.jpg

 

Link to comment
Share on other sites

  • 3 weeks later...
23 hours ago, Mikes said:

 

Does anyone know what the legal usage is for these creations? Like, can you use them as book covers for instance?

 

And with a full sub you can!  Guess what I'm doing this weekend?

 

grid_0.png

 

grid_0.png

 

 

 

 

Link to comment
Share on other sites

Amazing images, its insane how good they've got the algorithms to these days.

 

Regarding copyright, I watched a good video about the legality issues with NFTs yesterday, and part of the problem is the issue of copyright for AI generated images (which have plenty of associated NFTs). A court in the US ruled that copyright only applies if a human creates the art, so AI art even with human prompts can't be copyrighted by anyone. This obviously creates a bunch of legal issues for owners and creators of NFTs that use algorithms to create art, as legally they may not actually have any ability to prevent other people using it. Amusingly the whole issue hit the big time in a case you may have heard of, where a monkey took a selfie and the camera owner tried to claim copyright. PETA waded in to assert the monkey's rights but the court found that neither the monkey nor the camera owner held copyright as a human didn't take the photo. It'll certainly be interesting looking forward as to whether laws will change to allow AI generated art any modicum of copyright to people that use it. My thoughts are the most legit claims would be for the AI's creators, artists/works who taught the AI, and maybe the end user if they had enough creative input into the process.

Link to comment
Share on other sites

You can influence an AI image based on a real one by referencing it in the prompt (hyperlink) - with Midjourney you can upload your image to discord and then use the URL that generates. I find it a bit hit and miss whether it's doing anything though, and I've certainly not been able to tweak a real image by for example changing the sky or adding people. 

 

I've got DALLE2 now as well so I'll play with that later. So far it's much more realistic than MJ but it seems quite restricted as well. MJ produces much more creative and arguably interesting results, but DALLE certainly can do the wow factor if you get the prompts right. 

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue. Use of this website is subject to our Privacy Policy, Terms of Use, and Guidelines.