Jump to content

Text to Image Machine Learning


Sociotard

Recommended Posts

I've been binging on this topic and I want to start a thread to talk about it. For an article: OpenAI’s DALL-E 2 is a new illustration of AI bias - Vox.

 

These programs are similar in a way to Jukebox for music or GPT3 for text. They don't photobash existing pictures, but they draw on patterns they've learned to associate with certain words. And they're fun! just pop in a few words and get instant gratification.

 

I don't have access to Dall-E, but Midjourney just went to open beta, and I've been having fun with that.

 

Questions: Will this be impactful at all? I saw a few musicians 'collaborate' with Jukebox, taking inspiration from Jukebox output but fixing and improving things. And that was when it was new and fun; I haven't heard of anyone using it recently. It just kind of died. Is that what we should expect here? I've already seen videos of artists drawing inspiration from the machine's output, but still needing to put in a lot of work. 


Iron Man selling george foreman grills in the style of Jack Kirby

Sociotard_iron_man_selling_george_foreman_grills_in_the_style_o_157f144e-4459-4cb5-9e05-3c09915da4ff.png.e0ce25cb924c6aa8cff6210db9f580d2.png

 

Lines at Comic Con in the style of War Photography

Sociotard_lines_at_Comic_Con_in_the_style_of_War_Photography_8beb2aa1-9845-4999-995b-8368c772b65e.png.906d3d88be0ecad6fa40a84b1efb4c8b.png

Link to comment
Share on other sites

I'm still having fun. Any requests?

 

Playing with what style you ask for is great, but there are obvious holes. I tried asking for some in "precious moments" and it had no idea. Same for Ringerike Norse art.

 

Anyway, here is a double quarter pounder cheeseburger in the style of H R Giger. As with quite a few of these, I could see using it as a launch point for photobashing, but not an endpoint. the middle has a piece that looks too much like a bun, and the meat could use more beef texture. Also, there is no cheese. I asked for a cheeseburger.

 

Sociotard_double_quarter_pounder_cheeseb

 

Same for "Zen Geometry in Celtic style". The symmetry needs corrected.

Sociotard_zen_geometry_in_celtic_style_8

Link to comment
Share on other sites

some prompts are complicated by word associations. I wanted some fantasy images with giants, but using the word "giants" tended to get me creatures with football helmets, because some of the images captioned giants were of the New York Giants. I tried "titan", and the results were clearly influenced by the monsters in "attack on Titan"

Link to comment
Share on other sites

I find it oddly comforting that text-to-image AI is bad at hands. Really bad. Every novice artist I've ever listened to has complained about the difficulty of hands and feet. I see so many pictures on this site where the hands look like they belong on a gibbering horror. AI has the same problem we do. :)

 

Sociotard_xenomorph_in_the_style_of_Char

 

Sociotard_kim_kardashian_in_the_style_of

Link to comment
Share on other sites

This comic has been on my mind as I've been playing with MidJourney.

Strip-Les-specs-cest-du-code-650-finalen

I don't know about Dall-E, but I've already seen some codelike specialization creep into the prompts for Midjourney.

 

:: separates phrases 
-aspect lets you set an aspect ratio for the output
image URLs can be used as a prompt input. that isn't particularly codelike but it isn't intuitive either.

I just thought that was interesting.

 

Link to comment
Share on other sites

I wonder how far we can push this.  Would any of these work?

 

Calvin and Hobbes in the style of Erol Otus

Mad Max: Fury Road in the style of Bill Sienkiewicz

Hello Kitty in the style of Simon Bisley

Doctor Strange in the style of Picasso

Star Trek II: The Wrath of Khan in the style of Frank Miller

Avengers: Endgame in the style of ukiyo-e woodblock printing

 

The Gigerburger makes me hungry.

Link to comment
Share on other sites

Douglas Adams once wrote: "The results were more often surprising than they were accurate, but it was worth it for the times when they were both." That's how I feel at the moment. Giving it just the right prompt is an art, and the joe biden fight is not proving easy.

 

Sociotard_donald_trump_43fb2b8b-4bac-410

Link to comment
Share on other sites

1 hour ago, Old Man said:

I wonder how far we can push this.  Would any of these work?

 

Calvin and Hobbes in the style of Erol Otus

Mad Max: Fury Road in the style of Bill Sienkiewicz

Hello Kitty in the style of Simon Bisley

Doctor Strange in the style of Picasso

Star Trek II: The Wrath of Khan in the style of Frank Miller

Avengers: Endgame in the style of ukiyo-e woodblock printing

 

The Gigerburger makes me hungry.

 

Some did ... not.

 

I can see bits of pattern from Calvin and hobbes, but this is not Erol Otus at all

Sociotard_Calvin_and_Hobbes_in_the_style

 

Mad Max Fury Road in the style of Bill Sienkiewicz worked better

Sociotard_Mad_Max_Fury_Road_in_the_style

 

Hello Kitty in the style of Simon Bisley is a maybe

Sociotard_Hello_Kitty_in_the_style_of_Si

 

Dr. Strange by Picasso is okay

Sociotard_Doctor_Strange_in_the_style_of

 

These stills from Star Trek II do not capture Frank Miller

Sociotard_still_from_star_trek_ii_the_wrSociotard_still_from_star_trek_ii_the_wr

 

Not horrible, but the faces need cleaned up a lot.

Sociotard_still_from_Avengers_Endgame_20

 

Link to comment
Share on other sites

Lol the Erol Otus one is even more disturbing than Erol Otus' actual work.

 

I really like Lingerie Hello Kitty and her angry pink bomb friend. 

 

Dr. Strange looks like the Kardashian one.  Is Rob Liefeld actually Picasso?

 

The mesopotamian city one is beautiful, the spaceship in the background really sells it.

Link to comment
Share on other sites

1 hour ago, Pariah said:

How about "James Webb Telescope images in the style of Vincent Van Gogh"? 

 

1 hour ago, Sociotard said:

here is the interpretation of that photo of the Carina Nebula

Sociotard_in_the_style_of_vincent_van_go

 

That's actually quite nice. 

Link to comment
Share on other sites

5 hours ago, Old Man said:

I wonder how far we can push this.  Would any of these work?

 

We could go up to the felony level and say Nancy (of the newspaper strip with Sluggo) in the style of Alberto Vargas...

 

((merged) ------

 

1 hour ago, Sociotard said:

here is the interpretation of that photo of the Carina Nebula

Sociotard_in_the_style_of_vincent_van_go

 

That suggests optical alignment problems in the telescope that seem not to exist in real life.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Unfortunately, your content contains terms that we do not allow. Please edit your content to remove the highlighted words below.
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
  • Recently Browsing   0 members

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