Starting today, I’ll gradually explain, step by step, to those who are interested, how to properly write a prompt for generating images. This isn’t a 100% recipe for creating masterpieces, as there are many nuances, but it will make a lot of things clearer. We won’t dive into the jungle of programming or try to understand how neural networks turn words into pictures. I also won’t overwhelm you with the order of tokens in the prompt, as it’s not that critical anymore. But if you’re curious to learn more in the future, feel free to write to me, and we can discuss it in more detail. So, let’s get started!
Introduction
Modern neural networks like DALL·E, Midjourney, and Stable Diffusion can create images based on text descriptions. The key to getting the desired result is a well-crafted prompt. In other words, the better you describe the image in words, the more accurate and beautiful the final image will be. This article explains, in simple language, the basic principles of writing effective prompts, how to adapt them for different styles (e.g., realism, anime, pixel art, etc.), provides examples of good and bad prompts, and offers tips for improving phrasing (including word choice, detailing, and special commands). By following these tips, even beginners can master the art of crafting prompts and generate high-quality images that match their vision.
Basic Principles of Structuring Prompts
For a neural network to correctly interpret your request, the prompt text must be clearly structured. The neural network cannot read your mind, so it’s important to describe what you want to see in the image clearly and in detail.
Basic Principles of Structuring Prompts
Here are the core principles to help you craft an effective prompt:
Specify the main object or scene. Start with the primary subject or scene you want to depict. For example, instead of a vague “girl,” write more specifically: “young female wizard.” This sets a clear theme for the neural network. Avoid overly general phrases without details — a common beginner mistake.
Add actions or context. Describe what’s happening or where the object is located. For example: “a young female wizard sitting on a rock, reading an ancient spellbook in an abandoned castle.” This helps the neural network understand not only who or what is in the image but also what they’re doing and in what setting.
Use descriptive adjectives. The more details you provide about appearance, character, or atmosphere, the better. Include colors, shapes, sizes, emotions, quality, and other traits. For example, instead of “dog,” write “fluffy small brown dog.” These details make the image more precise and closer to your vision.
Clarify appearance details. For characters, describe their clothing, pose, facial expression, or unique features. For objects or creatures, specify color, texture, or shape. For example: “a young female wizard in a purple hooded cloak and leather outfit with ornaments, holding a book.” Details make the image vivid and reduce the chance of errors. The more detailed your description, the better the result.
Describe the background and environment. Specify what surrounds the main object: nature, an interior, or a cityscape. For example: “…in an abandoned castle, with magical particles floating around and a full moon in the background.” This sets the mood and context. The background can be specific (“pine forest at dawn”) or general (“dark blurred background with a glow”). Ensure the background complements the main object.
Define the style or artistic approach. Indicate the desired style of the image: photograph, pencil sketch, digital illustration, oil painting, 3D render, etc. A single word can significantly change the aesthetic. For example, “digital art” will produce a different result than “photograph.” If you want to mimic a specific artist’s style or genre, mention it. We’ll dive deeper into styles later, but at this stage, include words that reflect the desired style.
Specify the color scheme and lighting (optional). To set the mood, describe the lighting (“candlelit,” “neon light,” “soft dawn light”) and colors (“in warm golden tones,” “black-and-white”). For example: “…under dim moonlight, in cool blue-purple tones.” This helps create the desired atmosphere, whether bright and cheerful or dark and mysterious.
Ensure logic and clarity. Make sure the details don’t contradict each other. Phrase the prompt so it’s clear which attribute applies to which object. For example, “cat and dog, red and blue” might confuse the neural network — which is red, and which is blue? Instead, write: “red cat and blue dog sitting together.” Avoid ambiguity and overly complex sentences. Simplicity and clarity are key.
Prompt LanguageAlthough many services support various languages, it’s better to write prompts in English for more predictable results. This is because neural networks were primarily trained on English-language descriptions. In Midjourney or Stable Diffusion communities, almost all prompts are written in English. You can draft your description in your native language and then translate it into English. For example: “маленькая пушистая собака” → “a small fluffy dog.”
How It Works
By following these principles, you’re assembling a puzzle of words that the neural network turns into an image. Start with the foundation (who or what), then layer on details — like painting with words.
That’s enough for today to start experimenting with these tips. Next, we’ll explore how to choose words for different artistic styles.