Variations and AI-made Art, Indecisions, and Curators…

1

There’s a place (photography) where indecision is interesting, and this is what I’ll use today.

It’s not really about indecision, it’s “decision displacement”.

A photographer often takes many pictures of a scene, and then chooses one. We all know the contact sheets, with crosses for nays:

In the ending end, you get ONE portrait, ONE place: one photograph. It’s about “content curation”. You have to choose one.

2

But of course, some artists just don’t. They opt “not to choose”, they show a few pictures.

The simple idea of showing a “contact sheet” was already in itself a variation game. It puts the audience in the position of the curator: “What would YOU choose?”. Something else, probably…

3

I wrote an article somewhere about Depardon and Shore.

A/ The French photographer shows two pictures in a street in the city of Glasgow. A slightly different angle, different time of the day, different weather. You compare, you watch, and you wonder: why both? What do these picture offer? What do they show? What does the photographer want?

It’s a bit as if the photographer was talking to you. “You see this place? What should I do? This chimney and this crane, they’re cool, right?”.

B/ The American photographer says this too, but to himself: somewhere, there’s only one place from which he’d take a picture. And thus each of his photos are like “magnetic”, there’s a perfection in lines, light, energy…

So you see this picture, and not another one:

Stephen Shore. Image from Steel Town

4

In the history of Arts, some artists like to offer variations too. They turn around things, like Monet with haystacks or Picasso with tomato plants.

5

I’ve always loved pictures but I don’t know how to draw, thus I made plenty of photos, and I collected books – today I’m able to make cool images with prompts and Artificial Intelligence. The skills needed are all about words, how to use them and describe things to get pictures.

In this field, people are always seeking a kind of perfection. Crisp, detailed, perfect pictures. There is a huge catalog of examples at https://lexica.art/

There are programs that can “batch” pictures, so when I make castles, I make 1000 pictures, because styles are fun to explore. These are “combinations” of words and styles.

But with a single prompt, one can also makes plenty of images, and each one will be different from the others, like these towers:

The wind:

Robots:

6

So well, there’s a tool for the spirit here.

When you work on a project (at school, in your company, etc), do you come up with one result? A few variations? Plenty of variations? Who’s the curator? You, in your head, or the client/student/customer? When is it a mistake?

I realized, when I began to work for other people, that my idea of showing plenty of things to the collaborator is a mistake. Then often don’t want to choose.

But sometimes you have to let them choose all the same! Because they don’t know what they want? Because they need to be disturbed by the fact they will HAVE TO choose?

And, you’ll be surprised by their choice…

Thanks for reading!

7

You can watch Depardon’s book on YT:

Is AI-made Art, Art?

Is AI-made Art, Art?

When photography appeared, I suppose that painters asked the same question. Is photography an “Art”? The photographers don’t even know how to draw, right?

Well, it became an Art, and some painters began to USE photographs to get inspired, to remember compositions and light, etc.

AI-Artists needs other skills. They play with Python programs, “settings”, and of course the prompt, the “sentence” we use to summon the models.

To write good prompts, you must think about organizing the words. Still, also you need culture, and visual art vocabulary (including photo vocabulary), but also, like in other arts, a way to combine, to “link” things nobody thought to link before… You also need an eye to determine what’s banal, ugly or cool.

I will always stick to the idea that “A good picture is a picture that stops the eyes”. A good picture must do that in a world surrounded by pictures everywhere.

These days I play with “classic painters”, I make new paintings, and well, that’s it:

Of course, the pleasure is to combine styles, to create new styles:

You can get crap

You can get illustration

Maybe ideas for stories – “here’s the story of the little girl who invented snow friends:”

Little things:

Color ideas:

Composition ideas:

Cool mistakes!

The AI artist must also be a curator, he constantly has to select what should be shown or not.

Well, you can make your images, for free. Just follow this:

Thanks for reading!

Quick-Eyed Sky’s Collaborations

I invented “Quick-Eyed Sky” decades ago for a web page about instrumental and avant-garde music. I kept the name!

If I needed a logo, it would be this airship…

Today, it’s used for my YouTube channel about images…

https://www.youtube.com/channel/UCkYi6dzJ5emaY0tPGat3k9Q

oOOo

2 of my images were exposed in Berlin in September 2022

oOOo

I made the masks in this trailer:

oOOo

I made some special effects (forest, fair, turning structures) for the Weird Wolves video:

oOOo

I made the cover of a book released in November

oOOo

Patreon: https://www.patreon.com/Quick_Eyed_Sky – advanced tips & many prompts.

Artstation: https://quick_eyed_sky.artstation.com

Instagram: https://www.instagram.com/quick_eyed_sky

Blog: https://afrenchtoolbox.wordpress.com/ https://ko-fi.com/jeanpascalm

Thanks for reading!

Digit Lovers vs Artists

There’s a pattern in photography.

Two kinds of photographers. Those who take photos with their camera. And those who play for hours on their Mac, with apps, with digits, HDR, and I correct this and that.

I already wrote articles about those.

Take a car.

You can be the tech type. You want to open it and fix this and that. Or you want to drive it to the lake and breath fresh air.

Today I’m very busy making images with Artificial intelligence. I have an Artstation (here: https://quick_eyed_sky.artstation.com/ ) and a YouTube channel here: https://www.youtube.com/c/JPRobocat

Very soon, I noticed the same pattern.

To make images, you need a Google Colab, which is a Python program with a bunch of settings, like a machine, and you enter a “prompt”, which is a “phrase describing what you want to see”.

There are 2 camps.

  • Most humans are obsessed and focused on SETTINGS, they study these for weeks, very deeply, line by line. They’re obsessed with numbers, and digits.
  • The other camp is obsessed with the PROMPT, which is like poetry. Choice of words. The way you present them. Their weight. Etc.

It’s always the same pattern.

  • Camp 1 is very serious. They are logical. They are focused on their numbers. They want DETAILS, crunchy pictures, they want lines. Techs.
  • Camp 2 is very casual. They try things. They read poetry. They enter lyrics and try to find new artists. They want mood, light, and composition. Poets.

You know what? Both camps are OK. But I’m in the 2nd one. You knew that already, right?

Do you watch the beach and the mountain, or do you watch your car’s engine?

Thanks for reading!

Do you focus on the prompt or on the digits everywhere? What about “clamp_max”: 0.05? Should I double it? Or should I add “ominous sky” in the prompt? Where do you like to work?

“text_prompts”: {
“0”: [
“Greg Rutkowski, long butterfly airship in the summer sky, Artstation”
],
“100”: [
“This set of prompts starts at frame 100”,
“This prompt has weight five:5”
]
},
“image_prompts”: {},
“clip_guidance_scale”: 50000,
“tv_scale”: 0,
“range_scale”: 150,
“sat_scale”: 0,
“cutn_batches”: 4,
“max_frames”: 10000,
“interp_spline”: “Linear”,
“init_image”: null,
“init_scale”: 1000,
“skip_steps”: 0,
“frames_scale”: 1500,
“frames_skip_steps”: “60%”,
“perlin_init”: false,
“perlin_mode”: “mixed”,
“skip_augs”: false,
“randomize_class”: true,
“clip_denoised”: false,
“clamp_grad”: true,
“clamp_max”: 0.05,
“seed”: 2397292033,
“fuzzy_prompt”: false,
“rand_mag”: 0.05,
“eta”: 0.8,
“width”: 1600,
“height”: 832,
“diffusion_model”: “512x512_diffusion_uncond_finetune_008100”,
“use_secondary_model”: true,
“steps”: 350,
“diffusion_steps”: 700,
“ViTB32”: true,
“ViTB16”: true,
“ViTL14”: false,
“RN101”: false,
“RN50”: true,
“RN50x4”: false,
“RN50x16”: false,
“RN50x64”: false,
“cut_overview”: “[12]400+[4]600″,
“cut_innercut”: “[4]400+[12]600″,
“cut_ic_pow”: 1,
“cut_icgray_p”: “[0.2]400+[0]600″,
“key_frames”: true,
“angle”: “0:(0)”,
“zoom”: “0: (1), 10: (1.05)”,
“translation_x”: “0: (0)”,
“translation_y”: “0: (0)”,
“video_init_path”: “/content/training.mp4”,
“extract_nth_frame”: 2
}

How to make images with Artificial Intelligence with a few words, with ruDALL-E

Hi everyone! The “Prompt To Image” processes are blossoming everywhere on the web since Katherine Crowson presented the VQGAN+CLIP tool and made it public. This “Synthetic Imagery” (or GAN Art) was fantastic, but a bit difficult and slow to use.

You’ll find plenty of articles about this, and hundreds of “Google Colabs” with the code to play with. The result is often slow (about one hour to get an image).

The next step was to find similar tools (you enter a text, you get an image) on web pages like https://hypnogram.xyz, https://text2art.com/ or http://gaugan.org/gaugan2/ or the very easy https://www.wombo.art/ (have fun!).

There are tricks you quickly learn to use with each tool. Adding words to the prompt, like artists names or words like steampunk – here are bridges, a mantis, an owl, and for the first one “bird leather gold“:

Each site has its flaws, and one must use them to get things. For example, GauGAN2 is made for landscapes, so if you ask “Lake and forest” you get a realistic scenery. But if you ask “Totem” it’s lost, and there come the cool things:

The possibilities are infinite. Just give two words like “Airship Fire”:

Not what we expected, but good images, inspiring maybe if you write stories, poetry, or if you draw. Make 20 of them with automation and you’ll find a few great pictures.

oOOo

I made plenty of movies with these:

oOOOo

This year the Russians invented ruDALL-E ( https://rudalle.ru/en/demo ) and it’s different, more realistic, and MUCH FASTER than every other similar tools. It needs about 1-2 minutes to make one image.

The results are less “digital artist”, and much more realistic, because it’s trained on millions of photographies (an AI must be “trained”). This morning, today, I made a few dozens, like these 3:

Yessss possibilities are great. And you don’t have to write in Russian, they translate. Good.

oOOOo

This team made a BOT, which is on Telegram (yes, the app, it’s on your phone and your Mac, right?). You’ll find it on the page, it’s here: https://t.me/sber_rudalle_xl_bot

  • On this bot, you use the ruDALL-E Malevich (XL) Model, which is very powerful.
  • Each prompt gives you THREE images, you just have to save them on your computer, and it works on your phone too.
  • You have to prompt in Russian. Therefore you have to use a translation tool like Google Translate to invoke it.
  • If you find a good prompt, you can and must repeat it: each time you’ll get NEW images.

Here are images with the prompt “Airship in the mist”, which is “дирижабль в тумане”. I made 135 of the same prompt today. I’ll make a clip later. Here are 12 of them:

These are cool, right?

Here’s my YouTube channel with plenty of clips made with these: https://www.youtube.com/channel/UCkYi6dzJ5emaY0tPGat3k9Q

Have fun!

Prompts of Invocations? How to make pictures with words.

Prompts of Invocations? How to make pictures with words?

There are Google Colabs which create images from just a sentence.

One colab I use these days: https://colab.research.google.com/drive/1n_xrgKDlGQcCF6O-eL3NOd_x4NSqAUjK#scrollTo=TnMw4FrN6JeB

Choose a phrase, “Brazil Morning Landscape” or “Giant Robot in Taiga”, whatever. You can just ask “The wind”!. You can increase the size of the image to 500 (pixels – more will crash the program). I also tend to put “steps per image” to 20 (I’ll get more steps to “see” the machine work). In the menu bar on the top of the webpage, go to “Runtime”, then “Run All”. And oh, OK, you need 30 minutes to get the last image.

On another page, I invoked “The Wind” and got this. The AI clearly needs more invocations 🙂

You can “color” this by another word. Here’s the wind with…

  • A painter (Alex Colville, Pict.1).
  • A website (Artstation, Pict.2).
  • A tool used for video games (Unreal Engine, Pict.3).
  • A way of painting for the movies (Matte Painting, Pict.4).

oOOOo

Here are Picasso and Klee for the wind. Who else would you try?

oOOOo

The word Artstation is commonly used because this site is full of great fantasy artists, and it gives dramatic pictures. I just added mist, dragon, emerald, dawn or airships…

It’s interesting to see how the pictures are built. “Storm Circus Tent Artstation Matte Painting” begins with a cloud and finishes this way. Here are steps 20, 40, 400:

I put 10 little movies with attempts here: https://www.youtube.com/channel/UCkYi6dzJ5emaY0tPGat3k9Q

Have a nice day! Thanks for reading!

Sorts of Parameters

My last blog article was about how it was fun and funny to create images, creatures, things, worlds and landscapes with an AI interface.

Yesterday I made those – oh I know, there are many colors here 🙂

So I began to think about “what words should I choose?” to make the best results.

I have to add just one word to create an effect

Is “magical” useful for the first picture? “james-gurney-panoramic-entrance-of-magical-forest-full-or-fireflies”.

I can add a name like Dali or Giger, here James Gurney on the 2nd picture. I wonder if the word “detailed” is effective: “detailed, james gurney, orbital panoramic garden”

I can “ask” less color and some bokeh (the blurry aspect of what’s far) in the 3rd: “in black and white, bridge landscape in the city night, under the rain, bokeh”

Or I can describe precisely a frame (like for the 4th one: “artist’s tools on black paper with a lamp on a desk”

I found these tips on a page:

Landscapes and scenes where the focus is not a single object tend to be most realistic

Prompts can be long, detailed and specific: eg. ‘young sorcerer heiress of a cyberpunk corporation, with dragons on her side’

You can have multiple prompts in a single image. Divide each prompt with the | character

Try using ‘unreal engine’, ‘photorealistic’, ‘van gogh’ or any other widely known style or artist in your prompts

You can assign relative weights to prompts using this syntax: ‘green cloudy dog:50 | snowy:25 | christmas:25’. The absolute values are irrelevant – the weights are relative to each other.

You can assign a weight of 0 to a prompt to eliminate a concept from the image: eg. ‘christmas | santa: 0’

oOOOo

I also follow some Twitter accounts (including by the person who invented this). Try #vqganclip.

They play with words too. Some persons are so good they don’t give the key words used for their marvels. That’s life!

If I add “painting” it had some brushstroke effect. I have to find artists’ names to mimic their style.

Frazetta Axe Bird:

Roger Dean Futuristic:

chris foss black airship painting space:

oOOOo

This “How to” is great: https://minimaxir.com/2021/08/vqgan-clip/

Huge page with examples: https://softologyblog.wordpress.com/2021/06/10/text-to-image-summary/

Article about this AI: https://ml.berkeley.edu/blog/posts/clip-art/

https://learn.adafruit.com/generating-ai-art-with-vqgan-clip?view=all

(To be continued)

Pictures I made with an Artificial Intelligence

Pictures I made with an Artificial Intelligence, VQGAN+Clip. It’s very fun. It’s infinite.

oOOo

You can metal

You can bokeh city

You can bridge

You can room

You can architecture

You can Rozalski

You can Basquiat

You can abstract

You can everything!

Dreams & Nightmares images with VQGAN+Clip IA

It’s been a long time I’ve been that excited with a computer invention. I’m old enough to have seen (is this phrase English?) the birth of Apple II, Pong, Macintosh, the Internet (and the web), personal then laser printers, or… First Person Shooters!

My last “Oh waow” moment is the discovery of VQGAN+CLIP images. This artificial intelligence tool is available for everybody. You’ll find tutorials in articles or on YouTube.

Ex:https://medium.com/nightcafe-creator/vqgan-clip-tutorial-a411402cf3ad

The IA is trained to invent images from a line of words.

You have to search a little, but here’s a page with a list of pages to start :

https://www.reddit.com/user/Wiskkey/comments/p2j673/list_part_created_on_august_11_2021/

The sentence “Geometric glass city from the future at dusk” gives:

“Glowing river” gives:

This imagery is characteristic. One should not use flesh or human or animal because it brings you into the “uncanny valley” of monsters and teratology.

Ask Imgur: https://imgur.com/search?q=vqgan

Too much color, but also sometimes a great color or mood talent:

Woodland Witch of the Night:

  • There are parameters. The human words are seeds. This is a cool idea. Unlimited, right?
  • It’s a long process and I personally don’t have the patience (I made only one with a dolphin). But it’s a beginning!
  • Soon we’ll get high-res images of these in a second. And movies.
  • It can be a cool source of ideas, for painters and others.
  • There are SubReddits, like https://www.reddit.com/r/deepdream/
  • You can add “in the style of” in the text.

What words will you try?

Thanks for reading!