extern/easydiffusion
1
0
Fork 1
mirror of https://github.com/easydiffusion/easydiffusion.git synced 2025-06-20 18:08:00 +02:00
Code Issues Packages Projects Releases Wiki Activity

Updated Writing prompts (markdown)

SpecificKnot 2023-02-16 23:15:29 +00:00
parent 870aad7e72
commit 5b39c25e3b
1 changed files with 25 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
Writing-prompts.md

@ -1,66 +1,56 @@
## The Basics
* Keep your prompts short don't write full sentences. Separate different aspects of your image by commas:
Keep your prompts short don't write full sentences. Separate different aspects of your image by commas:
`girl on a swing, green grass, pink trousers`
`girl on a swing, green grass, pink trousers`
* Stable Diffusion only takes 75 "tokens" into account. A token can be an entire word, but some long words can also count as two or more tokens. Each digit of a number is a token, so `1920s` counts as 5 tokens.
Stable Diffusion only takes 75 "tokens" into account. A token can be an entire word, but some long words can also count as two or more tokens. Each digit of a number is a token, so `1920s` counts as 5 tokens.
* Due to the way Stable Diffusion was trained, there are a few words that often improve the quality of images, e.g. `highly detailed` or `trending on artstation`:
Due to the way Stable Diffusion was trained, there are a few words that often improve the quality of images, e.g. `highly detailed` or `trending on artstation`:
![image](https://user-images.githubusercontent.com/5852422/196815094-2a3d075e-a641-43b2-a773-26ebbd249fb7.png)
![image](https://user-images.githubusercontent.com/5852422/196815094-2a3d075e-a641-43b2-a773-26ebbd249fb7.png)
* You can use the names of art styles, artists or computer programs to influence the look of the rendered images. A wide choice of modifiers can be found in the _Image Modifiers_ section:
You can use the names of art styles, artists, or computer programs to influence the look of the rendered images. A wide choice of modifiers can be found in the _Image Modifiers_ section:
![image](https://user-images.githubusercontent.com/5852422/196817139-fe7f6aa2-b658-4170-be79-e02af7ec604b.png)
![image](https://user-images.githubusercontent.com/5852422/196817139-fe7f6aa2-b658-4170-be79-e02af7ec604b.png)
You can also put the modifiers directly into your prompt to get many different styles:
You can also put the modifiers directly into your prompt to get many different styles:
![image](https://user-images.githubusercontent.com/5852422/196818870-c44ca641-bc3e-45a9-91a5-c66a7b671d1f.png)
![image](https://user-images.githubusercontent.com/5852422/196818870-c44ca641-bc3e-45a9-91a5-c66a7b671d1f.png)
## Emphasizing parts of the prompt
* You can mark parts of the prompt as more or less important. When you put a part of the prompt in round brackets, it becomes more important. If you use square brackets, the enclosed part becomes less important. By using multiple brackets you can make things `((even more important))` or `[[[very unimportant]]]`:
You can mark parts of the prompt as more or less important. When you put a part of the prompt in round brackets, it becomes more important. If you use square brackets, the enclosed part becomes less important. By using multiple brackets, you can make things `((even more important))` or `[[[very unimportant]]]`:
![image](https://user-images.githubusercontent.com/5852422/196799015-7cfa13db-dffb-4c3e-82e8-786c8cd7b2af.png)
![image](https://user-images.githubusercontent.com/5852422/196799015-7cfa13db-dffb-4c3e-82e8-786c8cd7b2af.png)
In the above example, the word "pink" in the prompt `girl on a swing, green grass, pink trousers` made all of the girls clothes and the see-saw pink,
not only the trousers. By lowering the weight with three square brackets `girl on a swing, green grass, [[[pink trousers]]]`, the impact of the word
"pink" is reduced and the shirt and the see-saw are no longer pink.
In the above example, the word "pink" in the prompt `girl on a swing, green grass, pink trousers` made all of the girls clothes and the see-saw pink, not only the trousers. By lowering the weight with three square brackets `girl on a swing, green grass, [[[pink trousers]]]`, the impact of the word "pink" is reduced and the shirt and the see-saw are no longer pink.
* Instead of using brackets, you can also provide the importance ("weight") as numbers: `girl on a swing:1.2 green grass:1.0 pink trousers:0.8`
The first weight impacts the entire beginning of the prompt. In the above example, `girl on a swing` has a weight of 1.2, `green grass` has a
weight of 1.0 and `pink trousers` has a weight of 0.8.
The weight acts a separator, so you don't need a comma behind it. There *must* be a space character after the weight. If you use a comma or a bracket,
you will get a warning like `WARNING cuda:0 Warning: '2.0,' is not a value, are you missing a space?`
Instead of using brackets, you can also provide the importance ("weight") as numbers: `girl on a swing:1.2 green grass:1.0 pink trousers:0.8`
The first weight impacts the entire beginning of the prompt. In the above example, `girl on a swing` has a weight of 1.2, `green grass` has a weight of 1.0 and `pink trousers` has a weight of 0.8.
The weight acts a separator, so you don't need a comma behind it. There *must* be a space character after the weight. If you use a comma or a bracket, you will get a warning like `WARNING cuda:0 Warning: '2.0,' is not a value, are you missing a space?`
## Testing variations of prompts
* Use curly brackets in prompts to try different words, e.g. the prompt `man riding a {horse,motorcycle}` creates two rendering jobs: One for `man riding a horse` and one for `man riding a motorcycle`:
Use curly brackets in prompts to try different words, e.g. the prompt `man riding a {horse,motorcycle}` creates two rendering jobs: One for `man riding a horse` and one for `man riding a motorcycle`:
![image](https://user-images.githubusercontent.com/5852422/196795838-88dec248-dbbc-4681-b00f-c16444e80a73.png)
![image](https://user-images.githubusercontent.com/5852422/196795838-88dec248-dbbc-4681-b00f-c16444e80a73.png)
* **Prompt Matrix** - Separate your prompt with the `|` character to explore variations quickly. E.g. `girl holding a rose | illustration | cinematic lighting` creates four task combinations automatically:
**Prompt Matrix** - Separate your prompt with the `|` character to explore variations quickly. E.g. `girl holding a rose | illustration | cinematic lighting` creates four task combinations automatically:
* girl holding a rose
* girl holding a rose, illustration
* girl holding a rose, cinematic lighting
* girl holding a rose, illustration, cinematic lighting
## Negative prompts
* The negative prompt can be used to describe what you don't want to see in the image. The AI sometimes ignores parts of the negative prompt, especially if the prompt and negative prompt contain very similar terms. In other cases, the negative prompt has big impact on the composition of the image. For example, if you put "shoes" into the negative prompt, you might get a barefoot person - or you get an image where the feet are out of frame.
The negative prompt can be used to describe what you don't want to see in the image. The AI sometimes ignores parts of the negative prompt, especially if the prompt and negative prompt contain very similar terms. In other cases, the negative prompt has big impact on the composition of the image. For example, if you put "shoes" into the negative prompt, you might get a barefoot person - or you get an image where the feet are out of frame.
A useful negative prompt is:
A useful negative prompt is:
`Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ((((mutated hands and fingers)))), (((out of frame)))`
`Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ((((mutated hands and fingers)))), (((out of frame)))`
## Further reading
* [The PromptBook](https://openart.ai/promptbook) from OpenArt, a platform and community dedicated to AI-native content, and is written by members from their community.
[The PromptBook](https://openart.ai/promptbook) from OpenArt, a platform and community dedicated to AI-native content, and is written by members from their community.
* [Stable Diffusion Modifier Studies](https://proximacentaurib.notion.site/2b07d3195d5948c6a7e5836f9d535592?v=b5b75a67cc52483c9965cfc141f6f582)
[Stable Diffusion Modifier Studies](https://proximacentaurib.notion.site/2b07d3195d5948c6a7e5836f9d535592?v=b5b75a67cc52483c9965cfc141f6f582) This is a list of hundreds of different style modifiers.
This is a list of hundreds of different style modifiers
[lexica. Art](https://lexica.art/) This site is a search engine for prompt and image ideas. **Warning:** Some of the images on this website can be NSFW (adult-rated), please use it at your own discretion.
* [lexica.art](https://lexica.art/)
This site is a search engine for prompt and image ideas. **Warning:** Some of the images on this website can be NSFW (adult-rated), please use it at your own discretion.
* [Prompt Examples and Experiments](https://strikingloo.github.io/stable-diffusion-vs-dalle-2#prompt-examples-and-experiments)
[Prompt Examples and Experiments](https://strikingloo.github.io/stable-diffusion-vs-dalle-2#prompt-examples-and-experiments)