Stable Diffusion Prompt Guide

Not really sure what words to use with stable diffusion? This guide will help you find out which words work best for you - and which words don't work at all! Son you'll be creating your very of masterpieces simply through the power of the English language :)

#stablediffusion

Links:

[jonmichaelgalindo]

"Focused didn't make it look any more focused." Did you not see the intensely focused gaze the cat in the middle-right suddenly developed? :-O

[ianjohnson]

Great video!!!. Timestamps with my summary:
@1:05 Focused - bad result minor changes ?
@1:29 Sharp - bad result minor changes ?
@1:58 Painting - Yes. definately strong word
@2:17 Chalk Art - Yes. definately strong word
@2:36 Concept Art - Small - Medium changes towards concept art
@3:05 Trending on Artstation - Medium changes
@3:31 Canon M50 - Yes very strong, definately photographs
@4:10 Close-Up - Yes, strong
@4:31 Charcoal Drawing - Yes, very strong
@4:47 Intricate - Yes small - medium changes
@5:35 Composite Phrases - works as expected
@6:27 Word Order - works as expected
@7:56 Punctuation - Some minor changes unknown reasons
@9:57 Scale - Yes, higher values seem to increase "bloom" and "deep fried" like look/texture

[MirrorValley]

This is absolutely brilliant! I'm working on incorporating some of this stuff into my animations, and having these takes so much time off of my experimentation. Thank you!

[TheLoneVertex]

I haven't been doing stable diffusion stuff too long, but some of the ones I found that do give me some results are:
- Ink drawing with watercolor wash
- Ink drawing line art
- Coloring book line art
- Game art design sheet / Concept art design sheet (Sometimes these two can produce similar but distinct results, and sometimes very different)
- 3-point lighting
- 8k uhd
- Long Shot
- Low Angle / High Angle / Aerial View
- Time of day is early evening (Or any other time you wish to enter; this one can have some drastic effects on your lighting, and seems to weight heavily even when near the end of the prompt)
- Golden Hour (Can be used instead of time of day for very specific lighting conditions)
- Fish eye lens
- Pointillism
- Grayscale
- Pixel Art / Pixel Art Sprite (Both are good but very different from one-another; without "sprite" it seemed to favor a more high-detail close up portrait)

[ThatHobbyHunter]

Thank you very much for taking time to share your experience and filming the process!

[zoombapup]

I've noticed that "by <artistname>" is slightly different than "in the style of <artistname>", but for the most part collecting the correct artists in the right order to introduce the right feel is what gets me. Also, you can use terms to alter viewpoint. Things like "top down" or "aerial view" etc.

[chipcustomizer]

Thank you for doing this - very informative and great to see the results right away! Your work is appreciated!

[robertfinnie]

This is a great idea and way ti comparisons nerdy. Please do more of these.
I was curious as to how changing the prompt in subtle ways would change the output. Very encouraged to see that changing the word order would emphasise the characteristics of the order, this is interesting and makes sense, and is helpful and useful to know.
Utterly bemused and lost as to how punctuation has any influence though, but again useful to know and be aware of.
Thanks again dude. Keep it up.

[SyntheticVoices]

Loving this. Will save me a lot of time on trial and error.

[ThiagoGlady]

Please do more of these! I love this video so much!

[dixie_rekd]

I have no real idea what to use SD for, but its fun seeing into the "AI realm" and fascinating to see "the ghost in the machine" making something amazing, from a quazi random database of studied works and a sort of noise generated reality, the fact that even its creators really dont know exactly what it's going to create is astounding to me. Of course I've looked into how it does its thing (generating random noise and then clearing it up, every step, tuning it to be more like the prompt) but that just makes me even more impressed, slight variations in the original noise map, (the seed) and you end up with entirely different output. thats not even mentioning all the different SD models and safetensors available.

I've also been playing with novelAI (not just its image generation) and its ability to create and hold onto details and create a coherent narrative feels almost like an entire multiverse of stories waiting to be told; characters, places, and events, guided by the user, but ultimately created by random noise inside the ether, a language model that specializes in song and story. I sometimes "play" novelAI as a text adventure, exploring a world that has never existed with characters noone has ever even imagined, the AI building a path, a story, and an entire world. I'll sometimes tweak the path manually, other times playing along with the AIs narrative.. its really little more than a super powerful autofill feature, except it remembers what happened 2000 words ago, allowing it to build on its own work.

and ChatGPT too... I spent 30 minutes the other day just chitchatting with ChatGPT, I told it to roleplay as Data (from star trek) and it actually did a damn good job of the roleplay, answering questions about Datas time on the enterprise, his brother, his cat, etc. it played along wonderfully.

i dont know, maybe I'm just weird and overromaticizing it but to me AI is the biggest technological leap since the steam engine (yes I'm including the internet, and even the transistor.)
And so far, we havent even scratched the surface of AI, current models (language, vision, or otherwise) are just a drop in the ocean of what is to come.

[potatoface]

This is extremely useful (especially showing difference word order makes)

[darthvulture]

thanks, very helpful! interesting to see the differences just some small changes did

[scottlandis]

This was very informative and stimulating.

[scourgehh]

This is awesome. You should do more of these. Maybe one for humans, like full body. The Ai loves doing just headshots, so testing terms that help give full body images that arnt abominations could be great lol

[topguyz]

I wonder if the word "focused" just made the cat more focused, basically giving it smaller pupils. Maybe words like "in-focus" or "sharp" have more impact on the image.

[xreve_]

Love it! Got yourself a new follower! ♥

[TheAndyroo]

I got it to do medieval tapestries depicting described scenes, pretty cool!

[TheMrawesomest]

I feel like I learned a lot, and learned nothing at the same time. It is a fascinating technology. A lot of experimenting for sure, and tons of fun.

[joaomirandaartista]

beautiful art with the cats my friend =) !