As artificial intelligence (AI) systems like DALL-E, Midjourney, and Stable Diffusion explode in popularity for creating images from text prompts, more people want to learn how to get the most out of these powerful tools. Writing effective prompts is key to generating quality AI art that matches what you imagined. Follow these tips to craft better AI art prompts and unlock the capabilities of AI image generation.
Understand the AI Model Capabilities
The first step is researching the capabilities and limitations of the AI model you plan to use. For example, DALL-E excels at photorealistic scenes but may struggle with some types of abstract images. Midjourney tends to create more stylized, painting-like images. Make sure your prompt plays to the strengths of the model and avoids things it doesn’t handle well. It will be useful to consider examples of prompts posted on various marketplace and forums.
Use Clear, Unambiguous Language
Write prompts that are clear, simple, and unambiguous to reduce confusion. Use concrete language and avoid vague terms that can be interpreted differently. For example, “a tall building” is vague, but “a 50-story skyscraper” gives a clear picture. Stick to literal meanings rather than suggestive language. Describe the exact scene you want the AI to depict.
Lead with the Most Important Keywords
Front-load your prompt with the most important terms and objects you want included in the image. The AI will prioritize those elements first when generating. Put supplementary context and descriptive terms afterwards. For example, “A red rose on a piano keyboard” tells the AI to focus on the rose and keyboard first.
Use Relevant Style, Genre, and Medium Keywords
When applicable, include some keywords indicating the style, genre, medium or other qualities you want the image to have. For example, terms like “Impressionist style,” “watercolour painting,” or “32-bit pixel art” tell the AI what specific aesthetic or qualities you are going for.
Give Context and Setting Details
While you want to avoid being overly verbose, do give some contextual details and setting information when relevant. Scenes with some environmental information will look more natural than an object isolated in white space. Phrases like “a spaceship flying over a futuristic city” or “an octopus on the sea floor next to a sunken ship” give helpful context.
Spark the AI’s Creativity with Unique Requests
Part of the fun with AI image generation is prompting the software to create things we couldn’t picture on our own. Think outside the box and make some unusual, unexpected mashups and combinations to see what it comes up with. For example, try out “An avocado driving a tank” or “A cupcake wearing a top hat”. Just keep the keywords clear and give some setting details.
Edit and Refine Based on Results
Crafting the perfect prompt is an iterative process. If your output isn’t quite what you envisioned, tweak the prompt wording based on what worked well and what needs improvement. For example, add or remove descriptive terms, change the setting, or even swap out one of the main objects while keeping the rest of the prompt similar.
Use Comparisons for Abstract Concepts
When trying to depict more abstract concepts, creative comparisons can help ground things for the AI and lead to better results. For metaphors to work best, choose something visually tangible rather than just another abstract concept. For example, “Tranquility flowing like a river” gives a concrete visualization of an emotion.
Simplify Ambiguous Requests
If prompts with ambiguous or subjective qualities don’t generate what you hoped for, get more literal. Simplify down to the basic visual elements rather than trying to depict intangible concepts. You can convey themes through symbolism instead. For example, “the intricacies of quantum physics” is hard to visualize directly, but you could show “a scholar studying a blackboard full of complex equations.”
Provide Guidance on Size and Framing
Without further instructions, AI image generators will create compositions at random sizes and with unpredictable framing. Provide some guidance to better frame the key objects the way you envision. Phrases like “centred in the frame” or “taking up most of the image” can refine the composition. You can also specify a size for more consistency: “a tiny elephant on a table” or “a towering 30-foot wave”.
Consider Lighting, Angle and Other Qualities
Lighting, angle/orientation, depth of field, motion blur and numerous stylistic qualities can drastically change the look and feel of an image. Provide some direction by including words like “from below,” “blurred,” “sharp focus,” “brightly lit,” “fisheye lens,” etc. Add these after establishing the main elements, as too many complex instructions can confuse things.
Let the Image Evolve and Expand
Services like Midjourney introduce new dimensions of possibilities with features that let AI images grow, move, and transform. Once you generate a base image you like, use commands in subsequent prompts like “a scenic vista expands downward revealing more landscape details”. This keeps the composition you want while enlarging and evolving it. Other videos and cinemagraph features bring AI art to life.
Fine-Tune Faces with Portrait Descriptions
AI models still struggle to generate accurate faces without reference images. Yet describing distinctive facial features and expressions in prompts can produce some surprisingly realistic and evocative portraits. Elements like age, gender, hair colour/style, beard presence, distinctive nose shape or laughter lines help render better faces.
Curate Collections with Continuity
When generating multiple images for a series or collection, maintain continuity by keeping key prompt elements consistent throughout. For example, a “still life painting series of various musical instruments on a wooden table” ties them visually. Or generate comic panels or animation frames using recurring characters, poses and settings. Just vary some keywords like musical instrument type or character facial expressions.
Let the AI Surprise You
While thoughtfully crafted prompts are important for steering results, part of the magic of AI art generation is those delightfully unexpected bits of originality. After establishing the core elements, leave room for the AI to improvise details. Avoid being overly prescriptive. Enable some creative freedom and you’ll get images that manage to surprise even the prompt author.
The key is striking the right balance between providing enough detail and leaving some room for the AI to interpret. Keep these tips in mind as you experiment with prompting. Tracking what works for you and reiteratively editing prompts will lead to custom images that are truly one-of-a-kind AI creations. With thoughtful prompting and some creativity, you can unlock the vast potential of artificial intelligence for generating stunning AI art.