Beginner's Guide to Prompting for Image Generation
Last updated
Last updated
When crafting prompts for image generation, it's important to recognize that the process differs significantly from interacting with an AI chatbot. With chatbots, you can communicate using natural language, which large language models (LLMs) understand well. However, due to the underlying principles of image generation models, they have limited comprehension of natural language. Therefore, prompts for image generation should be written in a comma-separated list format (also known as the Danbooru Tag Format). Here is an example of correct and incorrect formats:
✅Correct (comma-separated format): 1girl, 25 years old, red hair, blue eyes, short skirt, smiling, facing_at_viewer, upper_body
❌Incorrect (natural language): A half-body photo of a 25-year-old girl with red hair, blue eyes, wearing a short skirt, and smiling at me.
While natural language prompts may sometimes produce images, complex prompts in natural language can lead to a loss of detail or even complete misinterpretation by the model, resulting in incorrect images.
Most image generation models are trained using tags (also known as prompts) from Danbooru. You can find a large number of effective tags on this page.
Set Category to General and Order to Count, then click “Search” to view popular tags. Try prompts like "1girl, long_hair, smile, open_mouth, blue_eyes, skirt"
You can also use the search box to find tag suggestions. For example, if you type "eyes," related tags will appear.
While most Danbooru tags are likely to work, compatibility depends on the tags and images used to train the specific image generation model. Feel free to experiment!
Positive Prompt
A positive prompt tells the image generation model what elements you want to see in the image, such as 1girl or 1boy, long hair or short hair, and so on.
Negative Prompt A negative prompt tells the model what elements to avoid in the image.
Quality-Enhancing Prompts
In addition to specifying elements, adding prompts that enhance image quality can improve the final result. Phrases like best quality, ultra-detailed, and high resolution can be useful. Don’t worry—our “Auto-Enhance Prompts” feature, when enabled, will automatically add general quality-enhancing prompts in the background. You can focus solely on the elements you want to include. If you'd like to enter quality-enhancing prompts manually, simply disable this feature.
Negative Quality Prompts
Negative prompts can also help prevent errors, such as missing fingers or toes. Anatomical errors can sometimes occur, but by using negative prompts like missing fingers or missing toes, you can reduce the chances of these mistakes. Similarly, enabling Auto-Enhance Prompts will automatically add common negative prompts in the background to prevent these errors.
Other very useful articles:
