Generating Images Methods and Techniques
Positive Prompt
Subject
The focus should be on what you want to see in the image. Many people make the mistake of not providing enough details about the subjects.
For instance, if we aim to generate an image of a woman, a novice might simply write:
A woman
While this may result in some decent images, it leaves too much room for interpretation. It's common for the face to get distorted, but there are ways to address this.
So, how do you envision the woman? Do you have specific keywords to describe her appearance? Is she blonde? What is wearing? Is it standing, posing or laying in the bed? And what's the background scene like?
As an example, let's describe her as:
a beautiful blonde woman, in a sexy red dress on a balcony.
By providing such specific details, we'll generate images that are more in line with our vision.
| Result |
|---|
 |
Medium
Medium refers to the material used in creating artwork. Some examples are illustration, oil painting, 3D rendering, and photography. The choice of medium significantly influences the style, as even a single keyword can profoundly alter it.
Now, let's introduce the keywords RAW and photography into the mix.
We've already described our scene:
a beautiful blonde woman, in a sexy red dress on a balcony.
And now, we specify the medium as RAW photography. With this addition, we observe a shift in the images from a semi-realistic painting style to a more photo-realistic style.
Style
The style of an image defines its artistic expression, ranging from impressionism to surrealism to pop art, and beyond.
Now, let's infuse the prompt with additional styles:
shadows, cinematic lighting, realistic reflections, and highly detailed.
So, we have our scene set:
a beautiful blonde woman, in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, and highly detailed.
With these additions, the scene takes on a more life-like tone in line with a photograph, delving into realms of modeling intertwined with a touch of intimacy.
Resolution
Resolution is crucial for determining the sharpness and detail level of an image. Let's enhance our prompt with the keywords: 4k and sharp focus.
Here's our updated prompt:
a beautiful blonde woman, in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, highly detailed.
And now, we specify the desired resolution as 4k with sharp focus...
Although the impact of specifying the resolution will vary from image to image, it’s always a good idea to include these keywords just in case...
| Result |
|---|
 |
Remarks
It's worth noting that the images generated are already quite impressive with just a few keywords added. When constructing a prompt, remember that more isn't always better. You don't necessarily need a plethora of keywords to yield high-quality images. Sometimes, a concise and focused prompt can produce excellent results.
Negative Prompt
Employing negative prompts offers another effective method to guide the image generation process. Instead of specifying what you desire, negative prompts detail what you wish to avoid. These exclusions can encompass not only objects but also styles and undesirable attributes.
For simplicity and universality, we’ll utilize a basic negative prompt that does not alter the style.
disfigured, deformed, ugly
Prompt Engineering
Keyword Weight
You have the option to adjust the significance of a keyword by altering its weight using the syntax (keyword: factor), where "factor" represents a numerical value. A factor less than 1 denotes reduced importance, while a factor greater than 1 signifies increased importance.
For instance, let's consider adjusting the weight of the keyword "breasts" in the following prompt:
a beautiful blonde woman with (big breasts:1.2), in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, highly detailed, sharp detail, 4k
By increasing the weight of breasts, you're more likely to generate images of the woman with an emphasis on the breasts, while decreasing it tends to yield images with less focus on them. While this tendency holds true on a statistical level, individual outcomes may vary.
This technique of adjusting keyword weights can be applied not only to subject keywords but also across all categories, including style and lighting.
| Result |
|---|
 |
Parentheses and Brackets
Another method to adjust the strength of keywords is through the use of () and []. When a keyword is enclosed within parentheses, such as (keyword), its strength is augmented by a factor of 1.1, which is equivalent to keyword:1.1. Conversely, enclosing a keyword within square brackets, like [keyword], diminishes its strength by a factor of 0.9, equivalent to (keyword:0.9).
Multiple parentheses or brackets can be used, akin to algebra, and their effects are multiplicative:
- (keyword) is identical to (keyword:1.1)
- ((keyword)) is equivalent to (keyword:1.21)
- (((keyword))) corresponds to (keyword:1.33)
Similarly, the impact of employing multiple brackets is as follows:
- [keyword] equals (keyword:0.9)
- [[keyword]] is equivalent to (keyword:0.81)
- [[[keyword]]] corresponds to (keyword:0.73)
Combining Keywords
You have the option to blend two keywords together, a process known as prompt scheduling. The syntax for this technique is:
[keyword1:keyword2:factor]
The factor parameter determines when the transition from keyword1 to keyword2 occurs, and it ranges between 0 and 1. For instance, consider the prompt:
a beautiful [brunette:blond:0.6] woman, in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, highly detailed, sharp detail, 4k
When employing this prompt for 20 sampling steps, it operates as follows:
Steps 1 to 12 feature the prompt:
a beautiful brunette woman, in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, highly detailed, sharp detail, 4k.
Steps 13 to 20 transition to the prompt:
a beautiful blonde woman, in a sexy red dress on a balcony, RAW photography, shadows, cinematic lighting, realistic reflections, highly detailed, sharp detail, 4k.
The factor (0.6) determines the timing of the keyword switch, occurring after 20 steps multiplied by 0.6, resulting in 20 steps.
| Result |
|---|
|
Ready to try it yourself?
Put these steps into practice now and generate your own images on sexy.ai!