Negative Prompts on NightCafe

Negative Prompts on NightCafe

Original article by creator @Weresl0th. Big thanks from the NightCafe team!

What is a Negative Prompt in AI?

Negative weighted prompts (also referred to as negative prompts) are a way to instruct the algorithm that you do not want the creation to be influenced by this concept and avoid including it in the resulting image. By including a negative prompt additional control of the resulting image may occur. Any word (not governed by the safety filters) can be added to a negative  prompt.


Negative prompts can be used to reduce:


  • Text, watermarks, signatures

  • Blurry results

  • Additional body parts in human subjects

  • Specific features or subjects (such as people, animals, objects)


This technique also be used in an advanced fashion, such as:

  1. Adding artist names to influence the result to be the opposite of their style
  2. Adding colors to reduce the overall color scheme of the result
  3. Emotional words to change the tone and composition (e.g. adding words like “horror, scary, frightening” to get a “happier” result)


Negative prompts can be used on the following algorithms on NightCafe:

  1. Stable Diffusion (SD 1.4, 1.5, SDXL 0.9, 1.0 and checkpoint models in those families)
  2. Coherent
  3. Artistic

Constraints to think about when using a negative prompt:

  1. Like all text-to-image  generation, results are inherently random and with negative prompts the image may still have features or subjects you included in the negative  prompt.
  1. Some concepts may be so inherently associated with the words in your non-negative (“positive”) prompt that the result may still include undesired features and elements.

How to use Negative Prompts on NightCafe

Using negative prompts differs between the Studio display  (for Stable Diffusion) versus the Coherent and Artistic display (referred to as the classic view)
  1. For Stable Diffusion, the advanced mode toggle must be set to on to access the advanced prompt editor, which allows for prompts to be set as with a negative weight
  2. For Artistic and Coherent, there is no prerequisite setting.

Setting the Advanced Mode toggle (for Stable Diffusion creations)

The advanced mode toggle is located at the top left (on Desktop) and top of the page (on Mobile) of the studio view.



To set it on, click or tap on the tag next to the words “Advanced mode” and check that it is now green with the toggle to the right. This will open up access to the advanced mode settings, including the advanced prompt editor.  

Adding a Negative Prompt to a Prompt

1. Scroll down to find the Advanced prompt editor and click on it:


2. You will be presented with the following, although it may differ in presentation between desktop and mobile. The screenshot is for desktop:


The advanced prompt editor has a host of features which are explained <elsewhere>.  This guide will focus on adding a negative prompt.

3. There are two different methods of adding a negative prompt to a creation.
  1. A default negative prompt which can be quickly added through a toggle that also can be edited
  2. Adding a prompt and setting it as negative

Adding the default negative prompt

Click or tap on the toggle so it is displayed in green with the toggle to the right:
​


Alternatively, you can click on “Let me tweak it” to automatically add it and begin editing it.  We’ll talk about the contents of the default negative prompt and the features/concepts it is meant to exclude later on in this guide.

Adding a custom negative prompt

Any prompt line in the advanced prompt editor can be made into a negative prompt. By adjusting the weight of a prompt line using the slider below the prompt line, and changing the displayed value to being less than zero, it will become a negative prompt. This applies to the Stable Diffusion, Artistic and Coherent algorithms.

To see the weight slider, at least two prompt lines are required in the creation. To add another prompt line, click or tap on “Add Another Prompt”.


Then you will see multiple prompt lines, with sliders to adjust the weight:


Add the words you want to one of the prompt lines, and adjust the slider to the left so the weight value is less than zero.  Here is an example:


Things to Note and Consider

  1. NightCafe will provide warnings about the number of prompts that are ideal.  This is for a broad range of creations, but experimentation may lead to results which you like.
  2. You can have both the default negative prompt and a custom negative prompt in the same creation.
  3. While experimentation can lead to interesting results, having only negative prompts will result in unusual, distorted or unexpected creations. This is especially true for the Coherent and Artistic algorithms.
  4. For the Artistic and Coherent algorithms, having negative weights which exceed the total positive weight may lead to unusual, distorted or unexpected creations.

Negative Prompt Concepts

You can type in any word or modifier into a negative prompt line, and set a weight to -1 for Artistic and Coherent, or as low as -3 for Stable Diffusion. Let’s review the default negative prompt to see an example of this:

ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft


Weight: -0.3


These can be broken down into the following ways to influence the algorithm (in no particular order):

  1. Aesthetic - generate a more aesthetically pleasing (pretty) image:  ugly, out of frame, blurry, blurred, grainy, draft, cut off, tiling
  1. Portraiture/Subject  -  produce a more average looking person:  poorly drawn feature, extra limbs, disfigured, deformed, bad anatomy
  2. Composition - create images that do not have characteristics found in many pieces of art: watermarks, signatures

What about the weight? Any value below zero will influence the algorithm to try to not create an image with those attributes.  Depending on other settings and the other prompts, a higher weight may work better than a lower weight but isn’t necessarily required.


Note: Artistic and Coherent are older algorithms. There are limitations in the training and method of how those generate images. Negative prompts will not result in images that are similar to most Stable Diffusion results, especially for creations featuring people or animals.

Tips on Negative Prompt Words/Modifiers:

  • Don’t use words like “no” - the inclusion of a word or modifier in a negative prompt implies you don’t want it

  • Use keywords instead of narrative descriptions.  Example: use “ugly” instead of “ugly person”

Advanced Techniques

  1. For Stable Diffusion, you can use parentheses around a word and include a weight value to further accentuate that you don’t want this. Example:  “signature, watermark, (people:1.5)” would tell the Algorithm you don’t want to see signatures and watermarks, and you want few or no people present (this adding an additional multiplier of 50% to the word people in the negative prompt)


    • Related Articles

    • How to hide prompts?

      Hide prompts for individual creations You can hide prompts directly from the creation page for individual creations. 1. From the main menu bar choose My creations page, then hover over the creation and click the three dots icon > Edit. Or open the ...

    • What are prompts and how to use them?

      What are prompts? Generative AI prompts are written or visual cues, words or sentences, that tell AI Art Generators what kind of image you want them to create. Prompt fundamentals In order to produce graphics with a specific style, the text prompts ...

    • Inpainting on NightCafe

      Original article by creator @Weresl0th. Big thanks from the NightCafe team! What is Inpainting in AI Image Generation? Inpainting is a technique of changing an image that involves filling in the missing or undesired parts of the image, or removing ...

    • Introducing Fine-Tuning on NightCafe

      Fine-tuning is a feature that offers a more personalized AI experience. This allows you to train a model to understand and recreate specific styles, faces, or objects. How to Get Started with Fine-Tuning Access My Models: PC: Find it in the main menu ...

    • DALL·E2

      What is DALL-E? The DALL·E 2 algorithm is an evolution of the original DALL-E by OpenAI. DALL-E kicked off the whole text-to-image scene by demonstrating that it was possible. DALL-E 2 brings a whole new level of coherence, resolution, and prompt ...