The prompt is the most important part of AI image generation. A well-crafted prompt can mean the difference between a mediocre output and a stunning masterpiece. This guide will teach you the principles and techniques of prompt engineering to help you get the most out of CrIAr's AI image generator.
The Anatomy of a Good Prompt
Effective AI image prompts typically follow a structured format. While you do not need to include all elements every time, understanding the components helps you craft better prompts:
- Subject: What or who is in the image? ("a woman," "a cat," "a fantasy castle")
- Description: Details about the subject ("wearing a blue dress," "sitting on a windowsill," "surrounded by clouds")
- Environment/Setting: Where does the scene take place? ("in a sunlit forest," "on a rainy city street")
- Art Style: What should it look like? ("oil painting," "digital art," "photograph," "anime style")
- Lighting: How is the scene lit? ("golden hour," "dramatic side lighting," "soft ambient light")
- Camera/Composition: How is the scene framed? ("close-up portrait," "wide-angle shot," "from above")
- Quality Tags: Quality modifiers ("masterpiece," "highly detailed," "8K resolution")
Prompting Styles by Model
For SD1.5 Models
SD1.5 responds well to comma-separated tags and quality modifiers:
"beautiful landscape, mountains, lake, sunset, golden hour, masterpiece, best quality, highly detailed, 4k"
For SDXL Models
SDXL handles more natural language and requires less quality tagging:
"A majestic mountain landscape with a pristine lake reflecting the orange and pink colors of sunset"
For FLUX Models
FLUX excels with detailed, descriptive natural language:
"A photorealistic mountain landscape captured during golden hour, featuring snow-capped peaks reflected in a crystal-clear alpine lake, with wildflowers in the foreground and wispy clouds in the sky"
For Illustrious/Pony
These models use booru-style tags for best results:
"1girl, solo, long blue hair, red eyes, school uniform, sitting, looking at viewer, smile, cherry blossoms, masterpiece, best quality"
Common Prompt Mistakes
- Too Vague: "a nice image" gives the AI nothing specific to work with. Be descriptive!
- Contradictory: "a photo of a painting" creates confusion. Pick one style.
- Too Long: Extremely long prompts can dilute important concepts. Focus on what matters most.
- Wrong Model Format: Using booru tags for FLUX or natural language for SD1.5 reduces quality.
- Ignoring Negatives: Not using negative prompts leaves you vulnerable to common artifacts.
Advanced Techniques
Prompt Weighting
You can emphasize or de-emphasize parts of your prompt using weights. Higher weight means the model pays more attention to that concept.
Prompt Ordering
In most models, concepts mentioned first in the prompt receive more attention. Put your most important elements at the beginning.
Iterative Refinement
Do not try to get the perfect image on the first try. Start with a basic prompt, see what the model generates, then refine based on the results. AI image generation is an iterative creative process.
Practice Makes Perfect
The best way to improve your prompting skills is to experiment. CrIAr's free plan gives you 20 images per day to practice with — use them to try different prompt structures, styles, and techniques. Over time, you will develop an intuition for what works best with each model.