How to properly write AI Art text prompts

[@gruberjens4354]

Thank you for this video! I tryed desperately to find out how to write prompts and all the sources were just junk... You cleared so much myth and BS up in a few minutes... It is just great! :-)

[@kemikemi756]

It's like feeding a baby - start with just a little bit and add with every spoon

[@ContemporaryCompendium]

Truth. I've been shoving an entire ladle into the baby and kept coming out with some hideous monstrosity.

[@MytMowse]

extra words and gobbledygook, punctuation marks, quotes " ", symbols such as { } [ ] in prompts is just added noise to the ai bot when it's making the images. The only punctuation mark midjourney understands is :: when using weights. Depending on what version you're using of midjourney, there is a prompt text limit, the bot will ignore extra words after the limit or add them in some way to the image (as noise), possibility making strange artifacts in your generations.

[@designertjp-utube]

Wow! At frames 0:28 I started crackin' up with your sample example of an _"Everything & the Kitchen Sink Prompt"_ It immediately took me back to 1962 when the Kindergaten Teacher said to her young kids class: _"Well guys, it sure is raining outside today. Thus no outdoor recess. Let's see who can draw a Ladybug sitting on top of a Frog's head trying to dodge his sticky tongue!"_ Now that's how you cut the crap out. Since I was such a young *Creative* *Artist* back then, I automatically drew in a wet dew lily pond leaf to mount the colorful subjects on.

[@dutchscrapper2317]

Thank you very much for this great explanation. Very useful.

[@kdzvocalcovers3516]

intricate detail?..more like adding every element (except what you asked for)including the kitchen sink.. ..after some investigation i seem to have found that complicated,extensive ,counter productive crappy prompts in,,,garbage out...negative prompts will bleed into and influence the positive prompt inadvertently and detrimentally...fact...great video as usual..Thanks Vlad!

[@midnightstories5695]

Very good Vlad some great info there !!!

[@Geekatplay]

thank you

[@stockpile2137]

I am new to ai art and found it fascinating, in a way it is a digital artist's introduction to computer programming. Is there a way I can learn how to right proper prompt text as organized like you? Do you have a course or any recommendations?

[@Geekatplay]

Yes, I am finishing 6 week course with Digital Art, www.digitalartlive.com about AI art, they maybe will have recordings available. You can contact organizers about it

[@wretepstube]

YES, this is very helpfull! 👍

[@dunkinzsoda]

great video, very informative, thank you.

[@Geekatplay]

Glad it was helpful!

[@HadesTimer]

great video, your mic seems a bit low. I know you are probably trying to prevent peaking. But I really have to turn you up and if you move away from the mic at all, I can barely hear you. Otherwise great video as always!

[@Geekatplay]

Thank you. I will check

[@kenmorris2858]

Many thanks for the tips.

[@Geekatplay]

Thank you for supporting

[@sYnDaEx]

Subbed. Very helpful, thanks.

[@Geekatplay]

thank you for your support!

[@kevinshenoy]

thank you 🙏😊

[@Geekatplay]

You're welcome 😊

[@Roughneck7712]

Good video. If you happen to get to it, I’d love to know your technique for captioning dataset images. There is almost no content out there discussing how to best caption dataset images when training textual inversions and/or models. Also, what are best practices when captioning for object training vs styles. Training embeddings for certain styles is giving me quite a bit of difficulty, and part of the problem I think I’m having is how I’m captioning my dataset images. Captioning for object training is relatively straightforward, but styles are not working well for me

[@mrbabyhugh]

19:07 just "8K UHD" seems to work for me. i would add "natural textures" too. I use Playground btw, Dif1.5

[@Geekatplay]

nice, i need check it again with new updates

[@AnthonySell]

As a photographer, the thing that annoys me most about the prompting strategies is that most people using terms like depth of field have no idea what they are actually stating. The depth of field has to do with how much of your image is in focus in a given image. Greater depth of field means a smaller f-stop (higher f-stop number), and a wider area from the camera that is in focus. A shallow depth of field is the effect you get with a wide aperture or a close-up image, where the background falls out of focus (bokeh), and the in-focus area is vary narrow. Frankly, after testing these terms in Midjourney, the system has no idea what the differences are between f-stops or shutter speeds, and simply using the term depth of field will introduce some bokeh and unfocused areas in the image, but not usually in ways that accurately respond to a given camera/setting/aperture, etc. Given that, you may as well just use the term bokeh, as it has fewer characters and will result in fewer tokens in your prompt. Less is more.

[@Geekatplay]

terminology used in AI prompts based on how model was used and what words was used to describe input data. Best way to find, use /describe in MidJourney on image that have strong dof, boken, blur ... rets and see what it will give as result, then you will know what model expected to be in the prompt.

[@AnthonySell]

@@Geekatplay Agreed, the people who created the model have demonstrated a remarkable level of ignorance on a number of useful distinctions. But no, the best way is to use the shorten command and show details, as this is connected to the language model, the describe function is not. The terms you get from a describe command will not give you predictable results at the moment.

[@Hades-wt6xu]

I've been experimenting with AI Art Rendering there's been a lot of trial and error in pictorial descriptions to get something that I kinda want but not what I was envisioning

[@Geekatplay]

different generators will use different CLIP models, that is leading to different results

[@designertjp-utube]

Wonderful Prompt Tutorial! I keep thinking you meant to type *Elf* *Princess,* instead of Elf Princes inside your Image creation prompt. If you purposely typed it plural (instead of a requested female) then you've just created/generated a cute Tranny Elt. Question 2 was, when you place a "no this or that" prompt, do we always use a "double dash" preceding it or can we get by with just one dash in front of the "no"?

[@Visual_Notes]

Totally interested in this subject, but why is this sound so bad?

[@dwainmorris7854]

It still has the same draw backs as Midjourney , you cant bring in your characters and designs and maintain the same consistency for like say comic books and portraits of real people

[@hablemos.un.poco.]

Using artificial intelligence to create images (for example starray) I have not found a method to prevent the artificial intelligence from modifying an image uploaded by me. Despite repeated attempts, with different methodologies, artificial intelligence has always altered the face of the subject of the photo I uploaded to which I only wanted to change the clothes and the landscape. What should I do to prevent the artificial intelligence from changing the face of the subject I indicated?

[@Geekatplay]

Stable Diffusion give you better options on how much you want AI influence image.

[@saraszpekor2546]

Can the images be chosen from my own photos?

[@Geekatplay]

yes

[@416Mordo]

I think 8k and 4k makes a difference on different ai generators

[@Geekatplay]

yes

[@devnull_]

"do and not" ==> do's and don'ts"

[@Geekatplay]

thank you

[@johnclapperton8211]

Interesting that you wanted a female "elf princess", but the typo "elf princes" produced a male-looking (but only one) elf prince.

[@Geekatplay]

Elf’s like an angels ;) no gender

[@devnull_]

Thanks! BTW your audio sounds like you were inside an air duct.

[@Geekatplay]

thank you

[@MichaelFlynn0]

sound is in and out, and your accent makes it really hard to follow. You need to slow down.

[@Geekatplay]

i will, thank you for sugestion

[@MichaelFlynn0]

@@Geekatplay - thanks, because the content is marvelous.

[@itllBuffGaming]

@@Geekatplay appreciate the content. I’m sure you got it fixed by now, but it sounds like your noise gate is set too high, causing the mic not to pick you up on your lower tones, just lower the gate, or increase your gain a bit. Could also move the mic closer. Great video! Take care!

[@Bigtwinman]

I agree with the others and struggled with the audio but seen value in the information. Same with the text examples. A empty screen dominated the view area but the text was not viewable for me. Thank you for taking the time to publish and share your techniques. Respectfully

[@swettyspaghtti]

Between the accent and the mic over compressed/cutting out this is tough to hear .

[@Geekatplay]

sorry about that, going to fix ... at least mic, accent little bit harder

[@sixfeetwonder_]

Lol don't write opposite prompts it's confusing. Next scene, retro futuristic city haha

[@Geekatplay]

confusion is best part ;)

[@sixfeetwonder_]

@@Geekatplay haha 😄 I was just teasing man. Good video. X

[@ИвайлоСтоянов-б3м]

The video is great but I am struggling with creating more complex stuff like a man which is sleeping and showing a vision from his dream. I am sure the AI can make it but I don't know how to prompt it.

[@dekompose]

That prompt was crazy lol

[@SteveBurg2001]

The simple prompts work the best, for sure. Another good video, thank you!

[@Geekatplay]

Thank you

[@josephstanski5180]

Thank you for the very fine lesson in crafting prompts.

[@jonj4419]

Thank you. Your content is fantastic. I appreciate the detailed way you go into explaining how prompts can and cannot function. It has helped my work a lot.

[@Geekatplay]

Thank you for your support

[@elijahheuerman392]

The really helpful thing I’m looking for is few things Can you word it in such a way to have to separate subjects? For example a elf princess with knight on horse back how do you write that out so the ai doesn’t mix match all subjects bend of all three? Could you for example say 3 separate subjects 1 “ elf princess “ 2 “ knight “ on 3 “ horse “ ? Another thing I would love to know is there a list of the websites it uses to search for art work to use ? Where is that list so I can pick my favorites style and artist I want to use. For example do any of the websites use comic book artists or only extremely famous air bush artists instead? If I can access the same site then I then could pick what I want. I got the point the first text was wrong since it mostly addresses effects over more different effects on top of more effects. Yet if used one of them then would that even work? For example lans flare from top right Lastly what is :: 2 you called this weight essentially do you mean this has priority over all other subjects if so then is this a way write that for it to understand better? Once more “ elf princess “ , ::2 “ knight “ ::1 “ horse “ ::0.5 Is there a common command list for the a.I. to better control it?

[@maitotechlab9035]

If you havent figured it out yet, you can just write down what you want, like the princess and the white horse with golden armor, sometime you need to go by trial and error for fine tunning. If you want, lets say, cartton like, or manga like, or anime or wathever just write it down. If you want some some specific ar style just write "by artist name". I found very usefull to use chat gpt to look out for styles and model, like just ask chatgpt, list 10 famous cartoon artists and them google them to fund what you like more, same as clothes, or colors, like i wanted to know about oriental dresses, i just asked chatgpt, list me traditional oriental dresses, and voila... Its amazing how fast you can do beutifull images using this tools. Also the :xx sets a weight for that setting, 1 is middle, 0 is minimun, 2 is max.

[@cryptfiend461]

No matter what i type everything has 2 heads😑

[@Geekatplay]

in SD use negative prompt 'multiple heads' in MJ --no doble heads

[@lbfaith]

Can you explain a little more what 35mm is for? I can’t grasp it. Is that how sharp it is?

[@1Know1tHurts]

It is about focal length of the lens.

[@alexnorth3393]

Very helpful!

[@nic-ori]

Thanks.

[@washingtonron25]

Can you tell me why I keep getting a message that says that version 4 is not compatible with stylize?

[@Geekatplay]

not all options available for version 4 (yet), try reduce stylization

[@washingtonron25]

@vladimir chopine Thank you. I did as you suggested, and it worked.