One of the secrets of writing effective prompts for stable diffusion is in the ability to provide as much detail as possible about what we want to create without confusing the machine. It is really all about the balance of detail, structure, and power keywords.
What is in This Post?
How to structure a good prompt? What words to use in each prompt section to get the best results? What are the powerful words that make everything beautiful? Where to get good prompt ideas from?
Structure
As you probably know there are tons of ways to write prompts. But some structures are better (much better) than others.
Words order matters and special characters can make a difference. Things you put in curly brackets are considered important. Descriptions placed in square brackets are less important. The machine gives more weight to words appearing at the beginning of the description and less to the ones that appear at the end. Also if the description is very long, then each word has less weight.
- { Subject Description }
If you are looking for photorealistic results, start your prompt with curly brackets and include the details about what exactly you want to create there. Since the subject is usually the most important thing about the generated images that would be the thing to start with.
However, in some cases, it would actually be smarter to start with the type of picture you want to create. That is true for scenarios in which the type of image is hard to create using ai and therefore you need to emphasize the importance of the specific image type/style by placing this in the front of the prompt.
This description itself should include the subject, the action he performs, and adjectives that describe him best. The subject can be a person, animal, creature, robot, etc. But you need to be specific about it. You can: write a military person standing on the street. But it is better to write: A military old man with a machine gun walking on an abandoned street.
Effective Adjectives:
Descriptive adjectives you use also make a difference, here are a few that are used often:
- Cute
- Adorable
- Joyful
- Colorful
- Bold
- Medieval
- Fantasy
- Energetic
- Magical
- Mysterious
- Depressing
- Grim
- Futuristic
- Epic
- Manmade
Environment Description:
- Indoor
- Outdoor
- In space
- Solid Color
- Type of Image
State the exact image type you want to create: realistic photograph, digital illustration, or comic book cover, here is a big list of options:
Image Types:
- Digital illustration
- Character reference sheet
- Comic Strip
- Comic Book Cover
- Movie Poster
- Blueprint
- Oil Painitng
- Matte painting
- Photograph
- Collage
- Concept Art
- Concept sheet
- Pixel Art
- Fantasy map
- Sketch
- Hologram
- Origami art
- Clay
- Plasticine
Illustration Types:
- 3D Illustration
- Low Poly
- Comic Book Illustration
- Cartoon Illustration
- Block illustration
- Anime
- Charcoal illustration
- Ink illustration
- Woodcut illustration
- Watercolor illustration
- Pencil Illustration
- Collage Illustration
- Acrylic Illustration
- Line Art
- Psychedelic Illustration
- Fashion Illustration
- Children's Book Illustration
- Caricature
- Chalk Illustration
- Graffiti
Photo Types:
- Polaroid
- Portrait
- Sattelite
- Art Style Related Keywords
Art style can refer to the exact artistic style you want to create and terms that may relate to that style. You can write a pencil illustration to define the style, but adding terms like sketch and cross-hatching may help the algorithm understand the exact direction you are going toward with your prompt.
Overall Style Related
- Steampunk
- Clockpunk
- Cyberpunk
- Dieselpunk
- Atompunk
- Rococopunk
- Steelpunk
- Stonepunk
- Oceanpunk
- Elfpunk
- Acidwave
- Weirdcore
- Cottagecore
- Dreamcore
- Vaporwave
- Baroque
- Film noir
- Boho
- Dadaism
- Cubism
- Expressionism
- Fauvism
- Futurism
- Impressionism
- Neo-Impressionism
- Post-Impressionism
- pop art
- Precisionism
- Rococo
- Surrealism
- Street Art
- Suprematism
- Art Deco
- Abstract Expressionism
- Classicism
- Baroque
- Art Nouveau
Pencil Drawing Related Terms
- Cross hatching
- Pointillisim
- Carbon Pencil
- Ink and Paint
- Art Inspirations
It is good to list places/artists to take inspiration from. If you list more than one artists to be inspired by, the algorithm will create some sort of combination between the artists which often will result in a completely new style driven by all the influencers you mentioned. You can mention specific names. You can also mention artistic platforms like Art Station, Dribble, Behance, etc…
General Inspiration
- Art Station
- National geographic
- Dribbble
- Behance
- Deviantart
- CG Society
Can Be inspired by:
- Animation Studios
- Painters and Illustrators
- Computer Games
- Fashion Designers
- Film Makers
- Camera, Shot, Render Related Information
At the end of the prompt, it is best to place the related information about lighting, camera angles, render style, resolution, the required level of detail, etc…
Overall Helpful
- 4K
- 8K
- 64K
- Detailed
- Highly Detailed
- High resolution
- Hyper detailed
- HDR
- UHD
- Professional
- Golden Ratio
Lighting
- Studio lighting
- Soft light
- Neon lighting
- Purple neon lighting
- Ambient light
- Ring light
- Volumetric light
- Natural light
- Sun light, Sunrays
- Sun rays coming through window
- Nostalgic lighting
Colors
- Fantasy vivid colors
- Vivid colors
- Bright colors
- Sepia
- Dark colors
- Pastel colors
- Monochromatic
- Black & White
- Color Splash
Render
- Octane render
- Cinematic
- Low Poly
- Isometric assets
- Unreal Engine
- Unity Engine
- Quantum wavetracing
- Polarizing filter
Camera Shot Type
- Long shot
- Closeup
- POV
- medium Shot
- Closeup
- Extreme Closeup
- Panoramic
Camera Lens
- EE 70mm
- 35mm
- 135mm+
- 300mm+
- 800 mm
- Short telephoto
- Super telephoto
- Medium telephoto
- Macro
- Wide angle
- Fish-eye
- Bokeh
- Sharp Focus
View
- Front
- Side
- Back
- High angle
- Low angle
- Overhead
Examples
In the next example, I will take a description for a dinosaur illustration and see how adding different styles and changing some other words in its prompt will affect the result. The random seed as well as all other settings will remain the same for all the images generated in this example. That is in order to see the influence the prompt itself has on the generation.
The base prompt I used was:
- Model: v1-5
- Seed: 2204335083
- Steps: 60
- Sampler: DDIM
Light and Color Changes
Type of Image
Art Style
Art Styles Combination
Online Tools that Help Build Good Prompts
There are online documents as well entire website dedicated to the art of good ai prompts.
Git Templates
This is a git document with a few templates and instructions on how to achieve certain types of imagery in stable diffusion.
A website with lots of beautiful ai images in different categories and prompts for them. When you click on the image you like, besides the exact prompt used you also can see the stable diffusion model used for it, the seed number, cfg scale, and sampling steps. Helping you to recreate that same image.
One more site where you can browse through beautiful images and get the prompts used to create them together with all the other setting values.
Now What?
Now it is time to create your own prompts and bring your ideas to life. It is important to note that technology is new and is changing massively all the time. Most likely that as time passes it will be much easier and much more intuitive to create effective descriptions that produce amazing results.
Stay tuned for new tutorials.
Shop Related Products:
- Like
- Digg
- Del
- Tumblr
- VKontakte
- Buffer
- Love This
- Odnoklassniki
- Meneame
- Blogger
- Amazon
- Yahoo Mail
- Gmail
- AOL
- Newsvine
- HackerNews
- Evernote
- MySpace
- Mail.ru
- Viadeo
- Line
- Comments
- Yummly
- SMS
- Viber
- Telegram
- Subscribe
- Skype
- Facebook Messenger
- Kakao
- LiveJournal
- Yammer
- Edgar
- Fintel
- Mix
- Instapaper
- Copy Link
Thurough article, thank you very much. I can’t help but wonder if your images would have come out differently if you had corrected the spacing between punctuation. Instead of using the base prompt, “digital illustration, {cute dinosaur },4K, detailed, fantasy vivid colors, sun light” used this as the base; “digital illustration, {cute dinosaur}, 4K, detailed, fantasy vivid colors, sun light.” The bold/underlined section in the second version shows correct spacing, which I personally have found, especially around curly brackets, can impact the outcome of the generated image. Just something that popped in my head when I read your base prompt and noticed that you had copy/pasted the prompt for each of the subsequent images generated. It made me curious as to how, or if, your images would have turned out any different.
Interesting, never noticed a major impact on the generated results that is caused by the spaces. I will test it! Thank you!