Stable diffusion prompt tutorial. NEW PROMPT BOOK released!

In this Stable diffusion tutorial we'll look at OpenArts newly released prompt book where they put together tips and tricks of prompting. We'll look through it, react and I'll tell you my views!
Prompt book - openart.ai/promptbook
Chat with me in our community discord: / discord
Support me on Patreon to get access to unique perks!
/ sebastiankamph
Ultimate Stable diffusion guide
• Stable diffusion tutor...
The Rise of AI Art: A Creative Revolution
• The Rise of AI Art - A...
Ultimate Animation guide in Stable diffusion
• Stable diffusion anima...
Dreambooth tutorial for Stable diffusion
• Dreambooth tutorial fo...
5 tricks you're not using
• Top 5 Stable diffusion...
Avoid these 7 mistakes
• Don't make these 7 mis...
How to fix live render preview:
• Stable diffusion gui m...
CHAPTERS
0:00 Introduction
0:23 The Prompt Book
1:15 Prompt format
3:18 Modifiers
3:32 Modifiers - Photography
6:27 Modifiers - Art mediums
7:27 Modifiers - Artists
10:39 Modifiers - Illustration
13:32 Modifiers - Emotions
14:42 Modifiers - Aesthetics
15:56 Magic words
18:38 Parameters
24:57 img2img / in- & outpainting
27:05 Prompt book showcase
29:29 Closing words

[@sebastiankamph]

The FREE Prompt styles I use here: www.patreon.com/posts/sebs-hilis-79649068

[@valscripted]

what makes me the most excited is that this isn't a topic from years ago. I can't find tons of detailed, experienced online guides. It's happening now, this year. And it's tech of the future, in a few year we might be smiling at those attempts we make today. And it's fun experimenting with it right now, watching these videos 30 min after they've been posted instead of 4 years ago.

[@sebastiankamph]

Haha, yeah, you're totally right there! I know when I've looked for guides on other software previously, it's always a couple of years old. I will probably have to make new tutorials forever to keep up with what's happening in AI :)

[@piorewrzece]

EXACTLY!

[@amafuji]

I know what you mean. Things are moving so fast when I see a post about SD from August I'm like "Wow that's ancient history. I didn't know stable diffusion existed back then."

[@OtherWorldExplorers]

I'm going to end up betraying my age here. But those of us who played the classic d&d back in the day, the old pen and paper have a one-up on everybody else who has not. Because we had to describe everything in exquisite detail to make sure our friends with their characters understood what we were trying to describe to them. Still there's a lot to be learned about the order of the text of the AI understands and what you're trying to achieve.

[@sebastiankamph]

Hah, I hear that! I remember we had to make printer copies of new d&d character sheets because we liked making new characters. Us being kids, those were hard to come by back then.

[@Xomage999]

As someone who likes both dnd and -booru sites from which a lot of models use tags from, this AI tech fits me like a glove. Super fun making avatars and character concepts. Hopefully the improvements to come are every bit as crazy as people think they'll be.

[@madsnylarsen]

with a telezoom you get a ton of compression in the photo, this means a very low perspective, the effect is in very far distance two people may look very equal in size, while if you stand close to person A then person B would seem super tiny in comparison :), it's a pretty cool effect, especially in a video where you move and zoom at the same time, you get those funky perspective compressions where the background scales up/down but the person stays the same size.

[@cliveagate]

Another awesome tutorial Seb, very clear and easy to follow 👍 - for sure there's a lot more that goes in to creating good AI Art & Videos than people imagine. For me it's exciting and i'm enjoying the ride 🤸🤸‍♂🤸‍♀

[@sebastiankamph]

Thanks a lot, Clive! Happy to have you along for the ride. I agree, it sure is exciting 🌟🤩

[@Ecker00]

6:10 A telefoto lens is for brining the background closer to the subject. So you see a smaller area of the background, and often increases the DOF effect.

[@theindexfinger3434]

yeah, telephoto is not just a zoom, it compresses the background behind your subject. You can do pretty cool effects,

[@leestrz4153]

Precisely. I am a professional photographer and it is super useful especially when there are elements in the bg that you want to exclude/include etc. It is also very good at foreshortening the subject and making light dynamic. Very easy to differentiate the background from the subject. This is all just for portraits. Lotta other useful stuff for products, wildlife etc but many of the same principles apply

[@criddyla696]

Superb, great overview of an excellent resource... thanks pal

[@Razorfiend]

This is an amazing resource, thank you so much for sharing!

[@sebastiankamph]

Thank you for the kind words 😊

[@beholder4465]

Tyvm sir appreciate your vids🙌🏻👑🙌🏻

[@sebastiankamph]

So nice of you!

[@SineEyed]

Fun Fact: Crepuscular rays are nifty in that they seemingly converge on a point in the sky not far above the clouds. This has been used by many flat earthers as evidence for a nearby sun. These rays only give the illusion of divergence, however. It is a trick of one's perspective which has it appear to be that way. The same rays, if viewed from another location somewhat orthogonal to where someone sees them diverge, will appear as they actually are - as all rays of light are that reach us from the sun - effectively parallel, each one to every other.. 🤓

[@sebastiankamph]

I loved this fun fact! Never stop being you! 🌟

[@SineEyed]

@@sebastiankamph no worries there - I'll be true to myself through all my days.. 📶💯🏁

[@Unlucky-Knight]

As an artist, this is the reason I like Stable more than other ai like Midjourney and DALL-E. It might take time, but if your an artist that knows alot of art terms, the results could be beautiful from Stable

[@woszkar]

Thank you for this info! My SD skills've just leveled up at least 10 times!

[@sebastiankamph]

Glad I could help! Keep leveling up those SD skills. If you create something cool, show us in our Discord! See channel description for link.

[@IcarusHollow]

pretty sure this was answered but liminal space is basically like endless or infinite looking spaces and rooms. it makes a lot of people uneasy. there was a video game based on it, can't remember the name. but thanks so much for these tutorials. idk why i waited so long to actually look up info on how things work instead of blindly making things, usually not very successfully. I learned a lot between this and a few of your other vids. So thanks much

[@FreihEitner]

I wish that I could give this 2 or 3 thumbs-ups! As a relative newbie to Stable Diffusion, thank you for this

[@sebastiankamph]

Glad you enjoyed it! And thank you very much for those kind words 😊

[@LouisGedo]

9:24 Soroyama ........... amazing artist. One of my favorites. I love looking through my large Soroyama print books every once in a while

[@sebastiankamph]

Yeah, they're all great! And to think they feel modern today and some were made 40 years ago. Happy to see a fellow art enthusiast here, Louis!

[@jinghuidong1121]

This video saves my life. Thank you so much sir.

[@sebastiankamph]

Happy to help!

[@pbjandahighfive]

Seems like an acceptable beginners guide although they are missing basically all of the more advanced stuff that gives you some real power like prompt emphasis weights, matrixes, scripts, textural inversion, loopback, x/y plots, prompt editing, xformers, ect., but still an alright beginners tip book.

[@sebastiankamph]

Agreed! I think their goal is more focused towards beginners and light users as their tool is a web-based one. More tailored towards ease of use.

[@muuuuuud]

Fun stuff! Pleasant presentation and concise info. Thanks! ^___^

[@sebastiankamph]

Glad you enjoyed it! Thank you for the kind comment 🌟

[@user-xe2ek1td1x]

Telephoto helps do portraits without the fish eye warping that a wide angle lense does

[@gabriellekamph]

Best intros ever! 🤩

[@TikiTim]

Telephoto lense will compress the background, making the background bigger in relation to the subject in the shot. If that works here then that would be useful.

[@sebastiankamph]

Thanks for the insight, greatly appreciated :)

[@sweetfry]

doesn't it also keep like...faces etc from looking flat or widened or whatever? remember watching a guide on the different effects lenses have on faces etc

[@fredmcveigh9877]

Thankyou .

[@gerza71]

I don't use artist name I can stand on my own two feet. I use styles like hyperrealistic, baroque, science fiction, fantasy and etc Liminal means threshold in Latin or between two spaces. example you are standing where a door is. Behind you is a hallway and in front is the bedroom. Sleep is a state between life and death.

[@marcthenarc868]

The issue I find with some of the keywords, is that, most likely, the training images were scrubbed from the Internet with Google keywords from which most of them could readily constitute a prompt. When it comes to camera settings, like Cannon or Nikon 2 FZ, not all images were consistently labeled as such and bias was introduced to illustrate such "features". And it's not like SD remembers "all the existing keywords of the Internet" and if I want to paint in the style of some unknown person, it will make a rough approximation of it according to its bank of names it collected.

[@fernandoz6329]

Awesome book , awesome work!

[@sebastiankamph]

Thank you kindly! 😊

[@manleycreations6466]

Telephoto lenses typically have larger subject separation from the camera which in turn often creates more bokeh. So I would assume the AI associates telephoto lenses with blurrier bokeh.

[@sebastiankamph]

Interesting! Thank you for that insight 🌟

[@hplovecraftmacncheese]

When you go to the online book, there is also a link at the bottom to download the pdf for free.

[@RandomPickles]

Youcan major stuff by flattening the image with a telephoto lens. a "zoom" is what you could call crappy digital zoom on a phone. The difference between a real wide lenses and a telephoto lenses is drastic for effect.

[@studio85amsterdam]

Thanks mate. I can not find any information related to de forum models. Any thoughts? Like ddim and the bunch.

[@infographie]

Excellent

[@sebastiankamph]

Thank you!

[@ipwnzuall2000times]

great, now we need one for all the parameters/settings

[@sebastiankamph]

Check my videos, I have ultimate guides on most stuff.

[@spearcy]

Great stuff!

[@sebastiankamph]

Thank you, Stephen! 🌟

[@henriquegonfer7750]

Nice material is a good point to have some references

[@sebastiankamph]

Glad you think so! And thank you for the kind words 😊💎

[@henriquegonfer7750]

@@sebastiankamph Do you just tried to run the ShivamShrirao Dreambooth that works on a VM of Linux and uses less then 10GB VRAM? I will love if anyone make a good tuto about it to work on 1.5. Thanks!

[@sebastiankamph]

I have not tried the Linux release no. Sounds interesting though.

[@THEJABTHEJAB]

Telephoto lens effect, if it's a large zoom, will remove all perspective from an image. For example with an 800mm lens if you zoom in from afar on a group of people they are all the same size. Much different than if you were taking a snapshot of one person with people behind them up close. Also with a very long zoom everything in a small area is in focus

[@m_art_ucci]

I was coming here to say this. One of the coolest ways I've seen this used is in Jaws and anime. Jaws: a woman entering the sea from the shore. Because everything looks so close to each other, it seems like she reaches the sea very quickly. Anime: I remember Evangelion has this. Shinji and some friends are walking down the street and they seem to be sliding down, instead of walking, because they don't change size (relative to the distance travelled), but they are moving.

[@THEJABTHEJAB]

@@m_art_ucci I recently made a telecentric lens too, it lets you eliminate or even REVERSE perspective,. It is one of the weirdest things I've seen

[@sebastiankamph]

Wow, thanks both of you! I learned something new today. Whatever else happens, I can fall back on that 🌟

[@m_art_ucci]

@@THEJABTHEJAB That's amazing! Can I see that anywhere?

[@m_art_ucci]

@@sebastiankamph The way I usually think about features and stuff like that is: if there is, someone uses it. Haha

[@rendez2k]

Great video! So terms such as trending on artstation - does SD have the ability to go and look whats on there and feed that in?!

[@lpnp9477]

No, the training data used when the model was created included that Metadata. For now it's more or less current but in 5 years it won't be unless you're using an updated model.

[@EliSpizzichino]

"Telephoto is a zoom... great" sure it is, but with a long camera lens the perspective is flat, this is the difference! "Crepuscolar" from the italian crepuscolare = twilight

[@sebastiankamph]

Thank you Eli for the insight! 🌟

[@createtrouble]

You would use a telephoto lens for portrait photography with extremely narrow focus depth

[@belld001]

Great work. If only you can make it into PDF then sell it. I would buy it since it is much easier to read.

[@edwardwilliams2564]

How does one go about making those comparison charts for the different samplers with varying steps and cfg counts?

[@lpnp9477]

In automatic1111 you can use prompt matrix or s/r. They have a while section on scripts in the wiki.

[@JarppaGuru]

just promt and it gives you image that was captioned same. if you ask more than was trained it mix or not mix depend what seed you get now. you only can make images what was trained...

[@tengentopka727]

I'mma start to sell prompts rather than focusing on my art career now XD. If you have more than 1 character use () to describe their characteristics so it won't apply to other characters. plus repeating words also adds weight.

[@burntgod7165]

I'm a novice. What the difference between Epic Diffusion and Stable Diffusion?

[@youngwolf0]

telephoto lenses are great for portraits, as well as forced perspective shots. not sure if that translates to sd but I hope it does.

[@sebastiankamph]

Thank you for the insight!

[@MasonVeil]

Telephoto is not a zoom. It flattens the image.

[@Collants]

Thanks a lot like allways!!! You have the best SD channel!!!

[@sebastiankamph]

Thank you for your neverending support, you absolute superstar you! 🌟

[@Blooper1980]

So at 30min video of "read the book"?

[@SD_Alias]

I find it is very hard to learn good prompting when you want to create more realistic Landscape, architecture or product photos. Unfortunately most content and suggested prompts are Anime, SciFi, Fantasy, Steampunk, superhero related stuff…

[@ethan-fel]

there's nothing about the negative on the book, when negative are very influential to the render

[@sweetfry]

telephoto lenses helps make images look less flattened out or something that a wide angle might do.

[@Theexplorographer]

Prompt writing IMO is best something that should be learned by experience and not by a book. Other than basic guidelines that is. Here is why. I think it is very important as humans of all different shapes and sizes that we learn to better communicate. What all this prompting is teaching us is that sometimes, though we see what we want in our heads, we do not communicate that well to others. Ultimately, we will learn to better express ourselves the more we prompt for what we want. I think this is the single biggest reward from AI image creation. Sure we get awesome images sometimes..but we are learning to better communicate not only with the bot but with others as well. It should also be worth noting that it is in my experience, especially with Stable Diffusion parameters in the program matter. Just changing the ratio can drastically change your results. Prompt creation is really important, but knowing all of the parameters and how they change what the bot comes up with is super important as well. All of this leads to more experimentation and the realization that mistakes are going to happen and that's okay! I read the prompt book and I highly suggest you give it a shot as well, but never stop your experimenting and pushing the bot in different ways outside of what the book tells you to. Thanks for the video!

[@SineEyed]

I don't think this is a very helpful perspective. At least with regards to prompting for Stable Diffusion. To craft better prompts, you need to understand what's going on under the hood. You need to know the kinds of text the model was trained with, so you can better tell it what it is going to understand. You need to know which words get picked up and tokenized, versus which words get discarded and effectively ignored. Being able to communicate well to other people something you have in mind is not a transferable skill in this domain. Humans have a vastly superior set of tools for interpreting what is being communicated to us, compared to an ai model trained on a specific set and limited context of words. I would agree with you 100% if we were talking about Disco Diffusion, but SD and DD are entirely different animals..

[@Theexplorographer]

@@SineEyed Perspective.

[@SineEyed]

@@Theexplorographer indeed; some being more useful than others..

[@sebastiankamph]

I think you both have interesting insights into this and I'm happy to see two different viewpoints in a constructive manner. I love the experimenting factor of it, and I have been forced to since there was basically zero info when I started. But if I had guides and instructions about what went under the hood when I started, I would have gladly accepted those as a starting base for my experimentation.

[@Theexplorographer]

@@SineEyed Okee... 🙈🙉🙊

[@earthequalsmissingcurvesqu9359]

there is only one magic word: WILDCARDS

[@Fingle]

Oof time to rewrite the book for 2.0

[@guns1inger]

More than once I have asked for a polaroid photo style, only to be given a photo of polaroid photographs....

[@sebastiankamph]

😂

[@kallamamran]

No! Face restoration is not good enough most of the time. It's often like adding a lower resolution face over a higher resolution render :/

[@sebastiankamph]

Did you change from the default face restorer though?

[@kallamamran]

@@sebastiankamph I have tried them all 😊Some make it more blurry, some make it another face. Nothing as good as Remini. If we'd have Remini in SD.... Wow! 😬

[@lpnp9477]

In my experience it does the opposite. A high res face on a lower res body. Face restore has some good applications, but I usually leave it off. I can get good results in less time with "detailed face" and upres the ones I like.

[@mikerhinos]

Telephoto is a big zoom, so I don't know why they differenciated telephoto and 800mm... Hajime Sorayama is a great artist for sexy chromed cyborg women :) Not sure this book is useful, when I see for example at 14:24 : the most important thing in these prompts are the last part and not their keyword :( Except maybe for the last one because we see a face so what's the point on emotions when you don't see any face ? :/

[@sebastiankamph]

Yeah, you have a good point. Would need something more indepth. But I think it's a great start that someone's doing something! I expect to see more stuff like this :)

[@dreamingflurry2729]

Bokeh? - I prefer bokken (that's a wooden training sword for Japanese martial arts, it resembles a katana)

[@maxinecaulfield2556]

Then I recommend you to fust juck off

[@VIpown3d]

First

[@sebastiankamph]

My man! Always the🌟

[@VIpown3d]

Someone needs to be the support in every team eh :D

[@CyborganicDreamscape]

20:22 euler-a 50 is way too many steps. 20-30 is best. 15 is sufficient most of the time.

[@sebastiankamph]

I agree! Many samplers however are great at 50, like klms, which is one of the most popular ones. euler_a and ddim can run much lower

[@Mikes-Code]

Why I always get mutated people... lol

[@lpnp9477]

Which sampler, how many steps? Are you using a negative prompt?

[@Mikes-Code]

@@lpnp9477 oh, its an old comment. Back when i was a noob 😅 Now i am already making tutorials on the topic 😁

[@lpnp9477]

@@Mikes-Code whoops, didn't see the 4 months part haha, be well

[@PedroRodriguez-zi9it]

Great content , too many ads though 👎

[@adidadidu]

I think that you should have prepared this video a bit more. It feels like you are doing all of this on the fly without any structure.

[@skywatcher2025]

I would recommened that if you use an artist in a prompt, make sure they aren't against that. It's not legally binding, but it's respectful.

[@sebastiankamph]

I would love the world to be a little more respectful to one another. Good insight, thank you.

[@joseluisrodriguez3027]

and one ad, and other ad, and other more, and on and on and on and on and....

[@Fiveash-Art]

How do you get Stable Diffusion to stop generating so many 'People of Color'? There has to be some negative prompt for this one because it's extremely annoying. It would be fine on occasion, but every time I let this thing go with a batch of 10 or 15, .. midway through it's all African Americans. 😂I don't need all that.

[@sebastiankamph]

You can prompt for the type of people you want, and weight that up with ( )

[@blender_wiki]

Good entertaining video but the book is Quite useless

[@ragnarlothbrok367]

Ai is still stupid as fuck and its just disappointing, it either ignores the half of the prompt or gives you a shit quality and it always can render some twisted hands and spines no matter how much 'extra limbs, limbs missing' you will put in the negative prompt, it just sucks

[@lpnp9477]

In the standard model yes. Use civitai for better models. Even if you aren't making porn, the porn centric models can make much better looking sfw images of humans. The training data just has so many more people than the standard model, so it's better at building images of them.

[@75190255508]

The black ants always win, just turn off the TV.