This guide helps people create pictures that look like the ones in their mind. Sometimes, people have an idea of what they want the AI to create, but they can't explain it well. This guide shows how to describe your idea in a clear way.
- There are five parts to describing the picture. First, you talk about what the picture is mainly about, like the main subject and what they are doing. Then, you can add details about how things look, like what they wear or how their face looks. After that, you describe how the subject is standing or sitting, where they are looking, and how the camera should look at them. Next, you talk about what's happening in the background of the picture. Finally, there are some extra things you can mention, like the kind of light or special angles, but we won't talk about those here.
- To show an example, this guide uses a picture of a woman doing something. We found this picture and want the AI to draw something like it. Pretend you have the same picture in your mind and try describing it. Then, you can use the machine to see what picture it creates based on your description.
It's like using your words to tell the machine what you want it to draw. You can try it too and see what pictures the machine can make!
- When testing a new prompt that you are not too sure about, try to run 4 images then change the keywords in the prompt accordingly. Here are the first 4 images:
The keywords that were used are as follows:
- ((beautiful Caucasian woman sitting on the floor reading, wearing a black pullover and black tights, knees and feet together, book on her thighs))
- (blonde short crop hair),((knees bent)), (focused look, concentrating)
- not looking at viewer, full body, ((top down view))
- wooden flooring
Immediately you can see that the camera angle is correct and so is the background. The pants do not look correct because I used ‘tights’. Her reading material is also noticeably missing. Otherwise the rest of the general elements are there.
Now lets change up some of the keywords and emphasis in the prompt. Here are the next 4 output images:
The keywords are now as follows:
- ((beautiful Caucasian woman sitting on the floor reading a book, wearing a black pullover and black yoga pants, knees and feet together, book on her thighs))
- (blonde short crop hair),((knees bent)), (concentrating on looking at her book)
- (not looking at viewer), full body, ((top down view))
- wooden flooring
The first image from this batch looks already very close to the desired composition of the stock image we started with. If we like this result we can iterate off this image by remaking it from this seed while changing up the keywords more, and removing the extra hand using inpainting.
In summary, providing as detailed a description as possible to the AI via your keywords will result in as close to an output image to what you originally had in mind. Try not to throw a whole manner of words to the prompt and hope something sticks - this can give the occasional stunning image, but it might rarely be close to what you had in mind to begin with.