Prompt Engineering, in its most basic form, is the process of crafting prompts using certain keywords and orders to achieve a desired image. Foocus has a built-in GPT-2 Engine that can expand prompts based on context clues and its own trained understanding of what the Image Generator is capable of.
Let us know what questions you have in the Discord
Foocus has a built-in GPT-2 Engine that can expand prompts based on context clues and its own trained understanding of what the Image Generator is capable of. This means that most technicalities are handled by Fooocus itself and not the end user, streamlining and simplifying the process of generating images.
For example, if you want to generate an image of a dragon, you can simply type:
A dragon, snow, moon
and Fooocus would expand the prompt as such:
a dragon, snow, moon, light, intricate, elegant, sharp focus, beautiful dynamic, highly detailed,very sleek, professional fine detail, cinematic, dramatic ambient bright colors, perfect, warm color,epic composition, striking, brave, attractive, elite, best, vivid, clear, coherent, advanced, creative, cute,artistic, trendy, cool, gorgeous, awesome
All of the adjectives and keywords which seem meaningless to the end user are actually used by the GPT-2 Engine to generate the image. This means that the end user can focus on the prompts and not the technicalities of the image generation process. However, there still may be times when you need to include or emphasize certain keywords in your prompts to get the desired result.
If you wish to emphasize a certain keyword in your prompt, you can use the following notations:
(keyword)- This will emphasize the keyword in the prompt- You can also add more parentheses to further emphasize the keyword. For example,
((keyword))will emphasize the keyword even more - Alternatively, you can add a colon and a number within the brackets to specify the emphasis level. For example,
(keyword:3)will emphasize the keyword with a level of 3
- You can also add more parentheses to further emphasize the keyword. For example,
[keyword]- This will de-emphasize the keyword in the prompt- You can also add more brackets to further dampen the keyword. For example,
[[keyword]]will dampen the keyword even more - However, the colon-and-number notation does not work with brackets. For example,
[keyword:0.75]will not work.
- You can also add more brackets to further dampen the keyword. For example,
Despite this documentation, F4 Services recommends the majority of users instead explore the array of different "styles" we offer with Fooocus, as it leads to much better results with prompting than the emphasis notations.
If there is anything the ender user is capable of doing to improve the quality of their prompts, it is the order of the words. As a general rule with all Stable Diffusion models, words at the beginning of a prompt are more emphasized than words at the end of a prompt. For example:
A cat, long hair, sitting on a stool, looking at the camera, a dog in the corner of the photo
Is less likely to include the dog in the corner of the photo than:
A cat, a dog in the corner of the photo, long hair, sitting on a stool, looking at the camera
Below we can compare the results of each image to get better visualizations of the differences in the prompts.
With this in mind, it is better to keep prompts shorter and more concise, as the GPT-2 Engine is more likely to include all of the details you need within its prompt expansion, and any stylizing you'd like can be found in the LORAs we offer.