Prompting Guide
What is a prompt?
A string of words, typically separated by commas, that describes how you want to transform your picture into an AI masterpiece. It is what uplevels your everyday photo into your BeFake!
What is prompt engineering?
It is the process of structuring words in a way that can be understood by AI. It’s a way of translating an image in your mind into instructions for a neural network. Remember that no neural network can read your mind so the more descriptive you are about what image you want to achieve, the better. For example, if you want to transform your photo into a Picasso-style image, if you just type “Picasso”, this leaves a lot of room for the neural network to guess what you want. Did you want to turn your face into Picasso’s or did you want your picture to look like cubism artwork? If you want the style of Picasso, a better prompt would be: “artwork by Pablo Picasso, cubism, masterpiece, 8k resolution” This prompt provides more detail and more direction to the neural network and is what the “engineering” piece of prompting is all about.
Will prompts from Midjourney or Automatic 1111 work on BeFake?
While there is some overlap, our pipeline is custom and highly fine-tuned for social-type posting, so a prompt you use on other platforms will yield different results than it will on BeFake. As described below, we also use different symbols for positive and negative prompts so read this guide carefully to improve your prompt results. As with any stable diffusion platform, perfecting custom prompting will likely require some trial and error. So experiment, have fun and don’t get discouraged!
What is the typical prompt structure?
Medium + Subject + Style
(e.g. “oil painting of a young woman…” instead of “young woman, oil painting…”)
In addition to Medium, Subject and Style, you can also add words related to the following categories: Artist, Resolution, Color, Lighting, Website
Medium
Medium is the material used to make the image. Some examples are illustration, oil painting, 3D rendering, and photograph. Medium has a strong effect on your image because one keyword alone can dramatically change the style.
Subject
For your BeFakes, which are typically pictures of you and your friends, a subject can often be omitted. Our model is trained to preserve some of the identity of the people in your photos while also giving you freedom to level up your image with AI! So unlike other stable diffusion models, you do not need to indicate the subject of your image since the subject is you. If you do want to change the subject (e.g. grumpy old man), then type in the subject you want the neural network to generate and BeFake will transform the face(s) in your image.
Style
The style refers to the artistic style of the image. Examples include cubism, impressionist, surrealist, pop art, etc.
Artist
Artist names are also often very effective modifiers. They allow you to achieve an exact style by using a particular artist as a reference. Advanced prompters even use multiple artist names to blend their styles and achieve stunning mashups.
Resolution
Resolution represents how sharp and detailed the image is. Examples are highly detailed or soft focus.
Color
You can control the overall color of the image by adding color keywords. The colors you type will appear as a tone in the overall image or in objects.
Lighting
As with all images, lighting is important. Lighting keywords can have a significant effect on how the image looks. Examples include cinematic lighting, dark, volumetric lighting.
Website
If you have done any prompting at all before BeFake, you will have heard of websites such Artstation and Deviant Art. These websites aggregate images of distinct genres and using them in a prompt helps steer your AI creation toward these styles.
Keep in mind that you don’t have to include keywords from all categories. The above list is simply a checklist of what could be used. Sometimes super simple prompts like “snowing” are enough to create AI magic!
Does the order matter?
The more important words should be at the beginning of the prompt.
Should I use punctuation?
Commas can be used to separate distinct concepts or keywords. You can combine disparate styles and add fun words for effect like “melting” or “funko pop”. To emphasize a word/set of words use the plus symbol: “+”. You can use up to four plus symbols “++++” to make sure a keyword is emphasized in your generated image. Use double parentheses “(())” for anything you do not want in your photo—this is described in more detail below.
What if I don’t want something in my image?
You can also instruct the neural network NOT to put things in your images. These “negative prompts” work with objects, styles and colors. To indicate something you do not want in your image, use double parentheses around each word or phrase, for example if you are creating an image of the ocean but do not want fish in your image, simply type: ((fish)). Our pipeline already applies common negative prompts such as ((extra limbs)), ((extra fingers)) and so on. Stable diffusion is still learning how to get body parts generated properly. Until it does, we will help you out!
What do brackets [] and angle brackets <> mean in your prompts?
Our bracketed style names (Styles) and locations (Wanderlust) are much longer prompts that we have abbreviated so you don’t have an insanely long prompt. Do not delete the bracketed style name if you want to achieve the style or location you selected from the gallery. Similarly, our angle bracketed descriptions are shortcut prompts for the faces in your pictures and should also not be deleted if you want best possible output on faces.
Can I use the same prompt on a photo and get different results?
Absolutely! And this is one of the most fun things about BeFake, namely the magic and mystery of what the neural network will generate from precisely the same instructions. So feel free to explore!
Concluding Tips
To yield results closer to what you’re imagining, think of things that would visually compliment your picture. For example, if you want a fantasy-style image, adding mist or fireflies helps and could be included in your prompt. For landscapes, describe what types of trees or scenery you specifically want to see. For buildings, do you want to see skysrapers or modern glass homes? These important details help define your image with respect to place and time and will better help you achieve what is in your mind’s eye. The key is to have fun and not get discouraged. You often have to play with and modify your prompts to learn and start getting closer to the images you want to generate. This is precisely why we offer a subscription tier so that you can prompt and BeFake to your heart’s content!