Text Prompts

Prompt Engineering

Prompt engineering is the process of structuring words that can be interpreted and understood by a text-to-image model. Think of it as the language you need to speak in order to tell an AI model what to draw.

Prompt Format

Start by asking a list of questions:

1. Do you want a photo or a painting?
2. What’s the subject of the photo?
3. What details do you want to add?
4. In a specific art style or photo type?

Your answers will make up the prompt. This is not an exhaustive list, but will help you get a good start. Always experiment - the more you test and iterate, the better results you’ll get!

Important Tips

• order matters: The order of words in your prompt is almost as important as the word choices themselves. Prioritize certain words by putting them closer to the start of your prompt.

• separate phrases with commas: Your prompt starts getting loooong as you move deeper into this process. It becomes hard to keep sentences well-structured. Separate keywords with commas will help keep the structure nice and readable by AI.

• modifiers: Modifiers can define the style, composition, or perspective of the image. There are magic words that are proven to boost the quality of the image. Expanding your modifier vocabulary will help you massively improve your prompts. For example -

  • photography: polaroid, tilt-shift, color-splash, thermal, 800mm

  • art mediums: graffiti, pencil, clay, wood, fabric

  • artists: Derek Gores, Alejandro Bursido, Makoto Shinkai, Chesley Bonestell

    ☞ Mix and match artist styles to tune up the creativity

  • illustration: origami, needle-felted, isometric, vector, caricature

  • emotion: romantic, energetic, grim, fear, cozy

  • aesthetics: acidwave, dreamcore, after hours, baroque, brutalism

  • magic words: HDR/UHD/4K, studio lighting, unreal engine, bokeh

• CFG values: CFG stands for classifier free guidance. The default value is 7. Lower number gives the AI more freedom to be creative, while higher number force it to stick to the prompt.

• step count: This parameter controls the number of denoising steps. The default step count is 25. Higher can be better but only to a certain degree. Higher also slows down the generation speed. It is usually recommended to stick with the default.

• seeds: The seed controls the initial noise. The default is random. This is the reason you get a different image with each generation when all other parameters are fixed. You can reuse the same seed if you want to reproduce some results or more consistency.

• negative prompt: Use negative prompt to tell the AI what not to include in the image. If you don’t want black and white, type in “black and white” in the negative prompt field (don’t type in “no black and white”).

For more advanced techniques, check out the prompt book.

Prompt Assistant

We built smart tools to help enhance prompts with ease. The prompt assistant is located right at the corner of your prompt box.

Quick Enhance

• Tap <Quick Enhance> to automatically enhance prompt

• Tap to <See more options>

• Or <Regenerate> to see even more.

Customize Prompts

• Tap customize prompts to reveal prompt enhancer options
• Select an option to enhance prompts tailored to your needs

Image to prompt

If your inspiration comes from an image, you can upload it to generate a text prompt and refer to the prompt for your creations. This can be found by going to home → image to prompt.

The Prompt Book

The Prompt Book is brought to you by OpenArt. It is the most complete guide to help you learn essential building blocks and all the techniques to master Stable Diffusion and other text-to-image models.

The secret for generating good images:

• A well-written prompt consisting of modifiers and a good sentence structure.
• Well-adjusted parameters. You can always use the default, but sometimes fine-tuned parameters can generate better results. This will strengthen your ability to bring your creative vision to life.

Get the Prompt Book →