- Prompt engineering involves carefully choosing which words to put first in your prompt. For example, if you want to emphasize “dark” or “beautiful” in your generated image, put those words near the beginning of the prompt. If they are too strong, you can try moving them later in the prompt to reduce their influence.
- When using negative prompts, the order also matters. Putting a negative prompt word early will prevent those elements from appearing more strongly.
- Separators like commas and parentheses can also be used to give words different weights. Parentheses increase a word’s weight, while brackets decrease it. Commas act as soft separators between concepts.
- Longer prompts are not necessarily better. Stable Diffusion models have a token limit, so prompts longer than 75 tokens may not fully register. Focus on crafting a clear, concise prompt with the most important details first.
In summary, carefully structuring a prompt and putting the most important words first is key to steering Stable Diffusion’s image generation. Experiment with prompt order and formatting to see what works best for your desired images.