DiffusionDraw Prompt Tutorial

The essence of AI painting lies in transforming input text into an image. This process is straightforward, but crafting effective descriptive text requires some skills. This tutorial introduces beginners to the use of text prompts to enhance the output of DiffusionDraw.

Ⅰ. Fundamental Concepts

Text prompts are divided into Positive Prompts and Negative Prompts.

Positive Prompts represent elements you want to see in the image, like a beautiful sunset or a cute puppy. Negative Prompts denote things you want to avoid, such as obstacles or unnecessary elements.

Currently, the model primarily supports English prompts. It's advisable to use English whenever possible.

More content in prompts doesn't necessarily mean better results. The more you write, the harder it is for AI to understand your intentions. Simplicity is often a good strategy.

Ⅱ . Writing Effective Prompts

1. Prompt Writing Styles

（1）Descriptive Style:

Use multiple words to express elements in the image. Separate each word with a comma and adjust weights using specific syntax. This is one of the most commonly used styles and will be emphasized in this guide.

For example: city on Mars, 8k, exploration, cinematic, Science fiction, cyberpunk, realistic

（2）Natural Language Style:

Natural language is what we use for regular communication. For instance, a girl with golden hair wearing a white dress and a flower crown is sitting under a laurel tree, with a lavender field around her.

Avoid overly complex grammar, as the language model might struggle to understand, leading to deviations in interpretation.

Due to the added complexity of natural language, use it cautiously as it might impact the model's interpretive abilities.

2. Universal Formula

When writing prompts, you can apply the following formula. You don't need to include all categories, select the most relevant ones for your needs.

Formula = Subject + Scene + Style + Quality + Perspective + Colors + Lighting + Negative Prompts

For example:

Prompt：beautiful aesthetic digital illustration of a relaxed panda surrounded by an endless forest of weed wlop and Julia Razumova, realistic, photorealistic, hyperrealistic, unreal engine, deviantArt, trending on artstation, artstation HQ

Negative Prompts：text, error, extra digit, fewer digits, cropped, worst quality, low quality,signature,  watermark

（1）Subject

The subject is the essence of the composition, a pivotal anchor for the image's content. When crafting an image, it's essential to provide a detailed and accurate description of the subject. This allows the model to comprehend our intentions effectively. For instance, if we aim to generate an image of a girl, a mere "1girl" won't suffice. Instead, we need to delve into specifics about her appearance, attire, actions, and other details. Is her hair long or short? What kind of clothes is she wearing? What is she doing? Only by providing such details can the model truly understand our vision and generate an image that aligns with our expectations.

（2）Scene

The description of the scene within the prompt holds great significance. The scene dictates the overall atmosphere, emotional expression, and visual impact of the image. It's crucial to clearly depict the environment in which the subject exists, along with surrounding objects. Otherwise, the model might randomly generate elements that don't align with the desired outcome. Additionally, for specific images like illustrations or logos, you can include terms like "simple background" or "white background" to avoid generating complex scenes.

（3）Style

Different artistic styles carry distinct characteristics and modes of expression. For instance, Fauvism emphasizes exaggerated depictions of color and shape. Impressionism focuses on capturing changes in light, shadow, and color. Surrealism accentuates the fusion of dreams and reality, while Dadaism emphasizes subversion and disruption of tradition.

Additionally, you can describe the technique used for the artwork, such as hand-drawn, oil painting, watercolor, photography, or 3D rendering. Hand-drawn pieces often feature unique textures and colors, while photography emphasizes composition and lighting.

Mentioning an artist's name is a powerful stylistic modifier. Using this allows the model to directly reference a specific artist's style for creation. For example, Muxia is a renowned artist known for his contemporary style, while Picasso's Cubist art breaks conventional perspective and visual norms by portraying objects from multiple angles and facets. Utilizing an artist's name as a modifier lets the model adopt their style without excessive adjustments.

（4）Quality

Image quality is a paramount factor. Employing high-quality vocabulary, such as "masterpiece" or "best quality," enhances image quality and allure. Such words facilitate the model's comprehension of your image requirements, leading to the generation of more lifelike, high-quality images.

（5）Perspective

Perspective adds unique visual effects and expressiveness to images. Different angles and viewpoints offer distinct visual experiences. Common perspectives include foreground, middle ground, background, close-ups of faces, and full-body shots.

（6）Color Palette and Tone

Colors contribute to varied visual styles and atmospheres. Specify colors clearly in your prompts, such as red, blue, or violet. Warm colors evoke warmth, enthusiasm, and intimacy, while cool colors create a sense of calmness, freshness, and distance. Neutral colors provide a stable background to balance the visual composition.

（7）Lighting

Lighting plays a crucial role in prompts, influencing the overall image effect and atmosphere. Light can be used to create various effects and moods, such as highlighting facial details or enhancing object textures and shadows. Common lighting options include natural light, sidelight, backlight, and moonlight.

（8）Negative Prompts

Negative prompts usually encompass elements or attributes you don't want to appear in the image. These could include low-quality images, unattractive styles, watermarks, logos, or content unsuitable for children (nsfw).

For instance, if you want to avoid an AI-generated hand that doesn't look good, you could use a negative prompt like "twisted hands" or "fused fingers."

Commonly used negative prompts include: text, error, extra digit, fewer digits, cropped, worst quality, low quality,signature,  watermark

You can also include elements you don't want to see. For instance, if you want to avoid nudity, weapons, blood, or disturbing elements, you can use negative prompts like "nsfw," "weapon," "blood," "guro."

Ⅲ. Advanced Prompt Techniques

1. Prompt Length

The length of your prompt should match the resolution and iteration count of your image. Mismatching might lead to unnatural results. Avoid overly short prompts, as they can affect image quality.

2. Word Order

The order of words in your prompt significantly impacts the generated image. By default, each word's weight is 1, diminishing from front to back. Placing a subject at the beginning will make it central, while placing a setting first will make the subject relatively smaller.

3. Descriptive Word Weights

AI uses word weights to selectively generate corresponding elements in the image. Assigning appropriate weights helps the model understand your artistic needs better.

To assign weights, write the descriptive word, followed by a colon and the weight value. For instance, (rose:1.4) means the weight of "rose" is adjusted to 1.4 times (default weight is 1).

Note that weights above 1.5 have a substantial impact on the image, and values below 0.1 might have minimal influence.

4. Descriptive Word Techniques

Use more adjectives/nouns and fewer verbs for accurate results. Separate words with commas, e.g., "beautiful girl, holding flowers."

Prefer combinations like "adjective + noun" or "verb + noun" over standalone words, e.g., "handsome boy, dribbling basketball" instead of "handsome, boy, dribbling, basketball."

For complex terms, describe them separately for accuracy. E.g., "blue, shimmering crystal ball" rather than "blue, shimmering, crystal ball."

5. AND Syntax

Use AND to combine multiple words for a mixed effect. Note that AND must be in uppercase.

AND syntax directly combines multiple descriptive words for AI to interpret in one go. It's effective for achieving blended effects.

Example: Multi-colored hair (color blending effect)

Prompt ：green hair :1.05 AND white hair:1

Write the desired hair colors, separated by uppercase AND, with colons indicating weights for color distribution control.

Ⅳ. Prompt Examples

Example 1: Two corgi dogs running on a grass field.

Prompt ：2 corgi dogs running on grass field

Negative Prompts： text, error, extra digit, fewer digits, cropped, worst quality, low quality,signature,  watermark

Example 2: City on Mars.

Prompt：city on Mars, 8k, exploration, cinematic, Science fiction, cyberpunk, realistic, aerial view, hyper detailed, moody cinematic epic concept art, realistic matte painting, hyper photorealistic

Negative Prompts： text, error, extra digit, fewer digits, cropped, worst quality, low quality,signature,  watermark

Example 3: Anime-style girl

Prompt：jeanne d'arc from fate grand order, 1girl, (best quality:1.2), 

(high detail:1.1), (full face:1.2),  (looking at viewer:1.2)

Negative Prompts： text, error, extra digit, fewer digits, cropped, worst quality, low quality,signature,  watermark