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)

Sea/Snow/Sky and their French friends

I opened a book about Proust and found this : “Le temps n’est pas passĂ© sur le hall du Grand HĂ´tel de Cabourg au bout duquel on voit, par la porte-fenĂŞtre, la mer”.

“Time has not passed on the hall of the Grand Hotel of Cabourg after which one sees, through the French door, the sea”.

Obviously, the author made a tracking shot for the eye, from the hall to the large window then the sea…

In French, “la mer” arrives deliciously at the end of the phrase, opening it to the vast sky. As you know, words have a genre in French, the sea is a she

I said to myself that “la mer” sounds opened and grand and clear, a bit unlike “the sea”, which brakes a lot with its “S” – “Sea” sounds to me like a solid string.

Then I thought about the snow. Snow sounds GREAT for fallen, thick snow. But when it flies from the sky in magic light meandering flakes, I prefer the French one : La neige !

Sky” is great for the sky. It sounds big and clear. The French word is “le ciel“… it’s more pale…

PĂ©pite is greater than nugget. L’Or is brighter than gold. But wood is good, it’s sounds like wood. We say “bois“, alright. Some other words are cool in both languages : l’acier (steel), both are solid and almost blazing, right?

 

Of course, this means nothing. I touch here the infinite, fractal and subtle differences between your native language and the learned one. I can get the words, but I can’t really get their radioactivity, or tiny ones, through movies and conversations.

What do I see on this picture? Curtains/Rideaux. Plates/Assiettes. Clouds/Nuages. Candles/Bougies.

Candle makes me see the flame. Bougie makes me feel the wax. Ahhh it’s complicated!!

 

Thanks for reading!

(and sorry for my bad English)

 

le-grand-hotel-cabourg-ver3-l-xlarge