â–²Note
- The order in which the prompts are placed is the priority, and the weights are decreasing from the front to the back.
- Do not pile up too many cues, the AI will not be able to remember, if the content is particularly large, you can appropriately increase the number of iterative steps.
- If Stable Diffusion is used, it is recommended that the token value of the prompts is not less than 75 and not more than 150 (too little effective information is not enough, and too many words will be forgotten).
- In Mid Journey, Stable Diffusion software, the prompts should be in English, and all punctuation in the prompts should be in English.
- Words, phrases and sentences are basically equivalent. For example, "1boy, Lake, swimming" and "A Boy swimming in lake" are semantically the same, but the use of phrases is more in line with the semantic conventions of Stable Diffusion. Conventions.
Prompts content:
1. Overall description of the picture
Usually we need to consider the overall quality of the picture first, which needs to cover the following four aspects:
Description of picture quality: it can be classified as high quality, medium quality or low quality, 2k, 4k HD resolution, etc.
Art style description: CG, secondary, 3D rendering, posters, real photos, oil painting, ink painting, etc..
Description of compositional shots: the proportion of characters in the picture, whether it is a bust or a full-body portrait, frontal, top view or side view, whether a camera lens is needed.
Lighting or tone description: natural light, spotlight, backlight? Bright, dark, cool or warm tones.
2. Subject
In the case of a person, for example, the description of the subject usually includes the following two aspects: features, pose.
Characteristics: Characteristics usually include facial features, age, skin, hair, clothing and other decorative details.
Pose: the dynamics of the subject, such as sitting, standing, running and so on.
3. Scene
The basic scene description usually includes the time, place, light, weather, details and other descriptions, which are often adjusted according to the actual needs of the screen.
4. Order of Prompts
The order of the prompts determines the order of the contents considered by the AI when it produces the picture in steps, and the weight of the later words will be relatively low. Therefore, important prompt words should be placed in the front.
When manually modifying the weights of the prompts, you can use parentheses to specify the multiplier of the weights.
For example, (red) means to increase the weight of the word by a factor of 1.1, and [red] means to decrease the weight of the word to 90.91% of the original weight. You can also use (red:1.5) to increase the weight of the word by a factor of 1.5, or (red:0.25) to reduce the weight of the word to 25% of its original weight.
It is important to note that it is recommended not to adjust the weights by more than 1.5.
