The so-called prompt engineering is fundamental for obtaining good results in generators like Stable Diffusion. When we talk about describing an image, it's actually about creating a structure with special characteristics and elements that respond to very subtle details, including position. The new "Stable Diffusion Prompt Book" published by OpenArt is an excellent starting point that will help you resolve doubts, and why not, polish your current skills when developing unique prompts.
AI image generation has exploded across the entire Web. Social networks became the ideal home to welcome this new wave of art, with some very marked trends (for example, Instagram adores Midjourney). Platforms and mobile applications continue to multiply, as does the number of users interested in the practice. And that inevitably leads us to the concept of prompt: Before generating an image, algorithms need a description... and there's a lot to learn here.
This is especially true in the case of Stable Diffusion. Its open source model makes it more flexible, and that means its prompts have a very wide range of action. At the same time, the learning curve is a bit more demanding, and it's not hard to fall into frustration with the first sets of images. Fortunately, the folks at OpenArt have published the "Stable Diffusion Prompt Book", a book that will help you absorb the basics of prompt engineering and optimize the quality of your results.
Design the best prompts with the Stable Diffusion Prompt Book
- Do you want a photo or a painting?
- What is the subject of the work? A person, an animal, a landscape?
- What kind of details do you want? What kind of light do you imagine for the scene? What is the environment?
- Are you looking for a specific style? 3D graphics? A poster? Oil on canvas?
After that comes the order of words, which is of great importance for any prompt. Sometimes, the algorithm ignores part of our description because it doesn't consider it as important as we do, but if we move those elements forward, the behavior of Stable Diffusion (and by extension, the result) changes completely.
The third section of the book focuses on modifiers, which allow us to alter the style, format, or perspective of the images. This is joined by a series of "magic words" that users have discovered in recent months, and that have an enormous impact.
Photography has its own category and presents a table of expressions to achieve certain effects. It's also very useful to enter the focal length and camera model in the prompt: For example, if you add something like "Nikon D850, 50mm", the algorithm will try to imitate that combination. Alternatively, you can use references to devices like GoPro, iPhone, drones, or even CCTV for that particular "closed circuit" touch.
Another section I'd like to mention is that of media. Its choice can drastically change the appearance of an image, even if we repeat most of the prompt or the seed. In some cases, the best is to follow the path of oil paintings or graffiti, but it's also possible to experiment with elements like wood or clay. If your plan is to create portraits, pencil drawings are excellent.
And to finish, artist names. If you look closely at some prompts on Lexica and OpenArt, you'll see that users repeat certain names to achieve a specific style. One of the most common combinations is that of Stanley Artgerm Lau, the Polish artist Greg Rutkowski, and Alphonse Mucha, but it would be a grave mistake not to leave that selection. In fact, the WebUI for the offline version of Stable Diffusion has a button that chooses artists at random, and it definitely deserves your attention.
Stable Diffusion Prompt Book is a creation of Julián Herrera, Musical Sleep, Bob Chernow, Coco Mao, and Mohamad Diab, whom we know much better for his fantastic project Public Prompts. The last update is from October 28, and you can download the PDF completely free, without restrictions or credentials. Don't miss it!
Official site and download: Click here
