April 24, 2023

Introducing advanced prompting and prompt history

Realm is the most robust social and creation app for casual AI generation. We're excited to introduce advanced prompting and prompt history - read more about how these tools will help you 100x your generation flow!

Introducing advanced prompting and prompt history

What's new

Advanced prompt settings give users more controls to bring their imagination to life and improve the generation quality.

Access advanced prompt settings in your creation toolbar! You can adjust the default settings to achieve better control of content generation. Realm supports all Stable Diffusion parameters through an easy-to-use interface.

Let’s breakdown the advanced prompting options:

Negative prompt: This tells the model, "don't show me this." So, if you don't want certain things to appear in the output image, you can use a negative prompt to tell the model what to avoid. It’s highly effective but often overlooked feature in prompting, which can help you attain outcomes that would require a significant amount of time if you were solely relying on adjusting the positive prompt.

Realm has already curated the default negative prompt for you to ensure everyone can create high quality generations without much effort. You can easily change this by tapping the Edit button for your own edit. Realm remembers your last used negative prompt. Tapping the refresh button next to the Edit button will reset it to the default.

Without any negative prompts

Seed: This is a special number that you can use to get the same output image every time (ex. seed 139). It's like having a secret code that unlocks a specific image, as long as you use the same code and settings.

Seed: 10

Utilizing the property that the same combination of seed and prompt produces identical images each time, we can employ various techniques to our benefit:

  1. Directing specific character features: By altering the prompt, we can control different attributes of a character, such as their emotional state, hair color, or skin color. However, the smaller the change, the higher the likelihood of success.
  2. Analyzing the impact of particular words: To understand the influence of a specific term in the prompt, we can modify and test it out with the same seed. It is a useful way to test the prompts - you only need to change a single word or phrase each time!
  3. Modifying style: If we like the arrangement of an image but are interested in observing it in a different style or with small alterations, we can use seeds as a basis for adjustments. This technique is great for portraits, landscapes, and scenery!
"katy perry, full body portrait, realistic portrait"
"professional product photo of a blue trench coat, floating suspended midair"
"a beautiful girl"

Resolution: This is the size or quality of the output image. The higher the resolution, the more detailed and clear the image will be.

Sampler: This is a tool that the model uses to create an image. There are different options to choose from, like different paintbrushes, and each one will give you a slightly different result...you can experiment with different samplers to see which one you like best!

The sole noteworthy dissimilarity between Euler a sampler and the other two can be observed in terms of the resulting images. When compared to DPM*, Euler a produces images with smoother colors and less defined edges, resulting in a more "dreamy" appearance. If you want this effect in your generated images, it is recommended to use Euler a!

Sampler: Euler Ancestral

Sampling steps: This is the number of times the model goes over the image to try to make it smoother and less noisy. The more steps there are, the more time it will take to create the image, but the final product will be cleaner and clearer.

CFG Scale: CFG refers to"Creativity vs. Prompt" scale. Decreasing its value grants the AI more liberty to showcase creativity, whereas elevating it compels the AI to adhere to the prompt more stringently.

It operates like a slider, allowing you to regulate the degree of resemblance between the generated image and the text prompt. If you wish the image to resemble the prompt more closely, you can slide it towards the right. Conversely, if you desire a distinctive image, you can shift it to the left.

CFG = 1
CFG = 15

Denoise strength (with a start image): This is the setting that controls how much of the original image the model keeps versus how much it replaces with its own generated pixels. A higher denoise strength means the model will replace more of the original image with its own, which can make the final image look more different from the original.

Give these new features a try in your generation flow today. Happy exploring!