Cracking the Code of AI Prompting
Prompt engineering is the art of "talking" to an AI model to get exactly the image you want. While it might seem like magic, it's actually about understanding the hierarchy of keywords and how the model interprets weight and emphasis.
The Standard Prompt Formula
A "perfect" prompt generally follows this structure:
[Subject] + [Action/Pose] + [Attire/Details] + [Environment/Background] + [Lighting/Mood] + [Art Style] + [Technical Quality Tags]
Example Breakdown:
"A cyberpunk woman standing on a rainy street in Neo-Tokyo, neon lights reflecting on puddles, voluminous cinematic lighting, photorealistic, 8k, highly detailed, masterpiece."
- Subject: Cyberpunk woman
- Action: Standing
- Environment: Rainy street in Neo-Tokyo
- Lighting: Neon reflections, cinematic lighting
- Technical: Photorealistic, 8k, masterpiece
Mastering Negative Prompts
In many models, the negative prompt is just as important as the positive one. It tells the AI what to avoid. Common negative tags include:
(worst quality, low quality:1.4), text, watermark, signature, blurry, deformed hands, extra fingers, bad anatomy, mutated, gross proportions.
The Secret of Weights
Most AI generators allow you to emphasize certain words using parentheses. For example, (red hair) will give red hair more importance than red hair. You can even use numbers: (red hair:1.3).
Pro Tip: Use Danbooru Tags for Anime
If you are generating anime art on DreamGen, using comma-separated tags instead of full sentences often works better. Most anime models are trained on Danbooru tags. Instead of "a girl with blue hair in a school," use 1girl, blue hair, school uniform, classroom. Learn more in our anime AI art pro tips.
