How to Use Negative Prompts in Midjourney

Tokenized AI by Christian Heidorn
18 Dec 202222:07

TLDRChristian Heidorn from Tokenized AI provides an insightful tutorial on negative prompting in Midjourney, a tool for generating images from textual prompts. He explains that negative prompting is a subtractive process that allows users to remove specific objects or concepts from their images. However, it's not as straightforward as it seems and requires a good understanding of how the tool interprets prompts. Heidorn clarifies that Midjourney does not support image editing like in-painting features found in other tools such as Dali 2 or Stable Diffusion. He demonstrates the use of the 'no' parameter and negative image text weights to remove elements from images, showing examples of their effectiveness and limitations. The tutorial also covers the impact of complexity on the effectiveness of negative prompts and concludes with a creative example of constructing a scene using both positive and negative text weights to achieve a desired image composition.

Takeaways

  • 📌 Negative prompting is a technique used to remove objects or concepts from generated images in Midjourney.
  • 📌 Unlike positive prompts that add content, negative prompts are subtractive, aiming to eliminate specific elements from the image.
  • 📌 Midjourney does not support in-painting or image editing like some other tools such as Dali 2 or stable diffusion.
  • 📌 The 'no' parameter in Midjourney is used for negative prompting, which is equivalent to applying a negative text weight to a concept.
  • 📌 Negative prompting is less effective with complex scenes and works best when trying to remove non-core elements from an image.
  • 📌 Using the 'no' parameter or negative text weights can lead to abstract results, especially as more complexity is added to the prompt.
  • 📌 To use negative prompting, one can either add 'no' before the concept to be removed or adjust text weights to decrease the prominence of an element.
  • 📌 Negative prompting requires trial and error, and results can vary, sometimes not achieving the desired outcome.
  • 📌 The effectiveness of negative prompting can be limited, and it's not intended for use in highly detailed or specific image editing.
  • 📌 Text weights can be used to construct entire scenes by emphasizing certain elements and using negative weights to remove others.
  • 📌 The process of negative prompting in Midjourney is not perfect and may require multiple iterations to achieve the desired result.

Q & A

  • What is negative prompting in the context of Midjourney?

    -Negative prompting is the process of removing something from your prompt, allowing you to eliminate elements from a specific scene or image that you've created or described.

  • How does negative prompting differ from the regular, additive process of writing prompts?

    -While writing prompts is additive, where you add content and transformation to your prompt, negative prompting is subtractive, enabling you to remove elements from the image.

  • Why can't Midjourney edit images like some other tools?

    -Midjourney cannot edit images due to its limitation as a tool that operates through Discord, which is not a web application, making it complex to implement image editing without making the process cumbersome.

  • What is the 'no parameter' and how is it used in negative prompting?

    -The 'no parameter' is used in negative prompting to specify objects or concepts that you want to remove from the generated images. It is equivalent to applying a negative text weight to a concept in the prompt.

  • How effective is negative prompting in complex scenes?

    -Negative prompting is less effective in complex scenes. It is better suited for removing minor elements or concepts rather than editing super specific details of an image.

  • What is the difference between using the 'no parameter' and negative text weights?

    -The 'no parameter' is a simpler way to remove a concept, equivalent to applying a negative text weight to that concept. Negative text weights offer more control and precision, allowing you to adjust the emphasis on different elements within the prompt.

  • Can you give an example of a successful use of negative prompting with text weights?

    -An example is creating a 'Rocky Mountain landscape in winter' with a buffalo herd, but without trees or snow. This is achieved by giving different weights to the concepts of 'buffalo herd', 'winter', and applying negative weights to 'trees' and 'snow'.

  • Why does the image become more abstract with each added complexity in negative prompting?

    -As more complexity is added, especially with negative prompts, the image becomes more abstract because the tool has to fit more constraints into the composition, which can lead to less natural and realistic scenes.

  • What is the main challenge when using negative prompting in Midjourney?

    -The main challenge is that negative prompting is not a perfect science and requires trial and error. It may not always work as expected, and users may need to experiment with different prompts to achieve the desired outcome.

  • Why might negative prompting not work well with core concepts of an image?

    -Negative prompting might not work well with core concepts because these concepts are central to the overall image interpretation. For example, trying to remove 'eyes' from a close-up of a beautiful woman would not work, as the concept of a woman inherently includes eyes.

  • How can one activate remix mode in Midjourney?

    -To activate remix mode, you need to select an existing image variation and then use the remix prompt feature. This allows you to work off an existing image and make specific changes to it.

  • What is the significance of using both positive and negative text weights in constructing scenes?

    -Using both positive and negative text weights allows for greater control over the final image, emphasizing certain elements while removing or de-emphasizing others, leading to a more tailored and specific outcome.

Outlines

00:00

🎨 Introduction to Negative Prompting in Mid Journey

Christian Heidorn from Tokenized AI introduces viewers to negative prompting, a technique used to remove objects or concepts from generated images. He explains that while Mid Journey is powerful, it lacks specific editing features like in-painting, which other tools support. Negative prompting is a subtractive process, contrasting with the additive nature of regular prompts. The video promises to demonstrate both the capabilities and limitations of negative prompting through various examples.

05:00

🕯️ Removing Candles from Birthday Cakes with Negative Prompting

The video provides a practical example of using negative prompting to remove candles from generated birthday cakes. Heidorn uses the 'no' parameter to exclude candles from the image, demonstrating the process of using the 'remix mode' to alter existing images. However, he points out that the technique has limitations, especially when dealing with complex images or concepts that are closely tied to the core idea of the prompt, such as trying to remove eyes from a close-up of a beautiful woman.

10:01

🌼 Refining Nature's Meadow Scene with Abstract Prompts

Heidorn discusses the effectiveness of negative prompting with abstract prompts, like 'Nature's Meadow,' and shows how to remove elements like flowers and birds while keeping the composition similar. He explains that as complexity increases, the results become more abstract and less realistic. The video illustrates how negative prompting can be used to refine scenes by removing unwanted elements while maintaining the overall composition.

15:02

🏞️ Crafting a Rocky Mountain Landscape with Text Weights

The video moves on to discuss an alternative method of negative prompting using text weights. Heidorn demonstrates how to construct a scene with a specific emphasis on a buffalo herd in a winter landscape without trees or snow. He shows that by assigning weights to different elements of the prompt, one can control the prominence and inclusion of various features in the generated image, allowing for a high degree of customization.

20:02

🔍 Conclusion on the Nuances of Negative Prompting

In conclusion, Heidorn emphasizes that negative prompting is not an exact science and requires trial and error. It's not as precise as in-painting in other tools but is a valuable tool within Mid Journey's capabilities. He encourages viewers to experiment with negative prompting and text weights to achieve their desired outcomes. The video ends with a call to action for viewers to like, subscribe, and look forward to future content.

Mindmap

Keywords

💡Negative Prompting

Negative prompting refers to the process of specifying what should not be included in the generated image by an AI system. In the context of the video, it is used to remove unwanted objects or concepts from images created by the AI software Midjourney. The technique is subtractive as opposed to the additive process of regular prompting, where elements are added to the prompt to influence the AI's output. An example from the script is the attempt to remove candles from a birthday cake image.

💡Midjourney

Midjourney is an AI tool used for generating images based on textual prompts. It operates through a Discord interface and is mentioned in the video as being limited in its ability to edit images directly, unlike some other tools that support in-painting features. The video focuses on how to use negative prompts within the constraints of Midjourney's capabilities.

💡No Parameter

The 'no' parameter is a specific syntax used in prompts to indicate that a certain element or concept should be excluded from the generated image. It is demonstrated in the video that adding 'no' before an object, like 'candles', to the prompt will attempt to remove that object from the AI-generated image in Midjourney.

💡Remix Mode

Remix mode is a feature within Midjourney that allows users to modify existing images generated by the AI. It is activated to work off an existing image, making minor changes to it. In the video, the host uses remix mode to remove candles from a birthday cake image by adding the 'no' parameter to the remix prompt.

💡Text Weights

Text weights are numerical values assigned to words or concepts in a prompt to control their importance or emphasis in the generated image. A positive weight increases the prominence of a feature, while a negative weight, indicated by a minus sign, reduces its presence. The video explains that the 'no' parameter is effectively a shorthand for a negative text weight.

💡In-Painting

In-painting is a feature supported by some AI tools that allows for the editing of specific parts of an image, such as removing or adding elements with a more precise control than negative prompting. The video mentions that Midjourney does not currently support in-painting, which is a limitation when compared to other tools like Dali 2 or Stable Diffusion.

💡Composition

In the context of the video, composition refers to the arrangement of elements within an image. When using remix mode in Midjourney, the composition of the original image is retained, but certain elements can be modified or removed based on the negative prompts given.

💡Abstract Prompt

An abstract prompt is a more open-ended and creative input that can lead to a variety of interpretations by the AI. The video uses the term to describe prompts that are less specific and more artistic, such as 'Nature's Meadow,' which can generate a wide range of images that are less tied to a single, concrete concept.

💡Conceptual Link

A conceptual link refers to the inherent connection between certain elements and the overall theme or concept of an image. The video explains that negative prompting is less effective when trying to remove core components of a concept, such as eyes from a close-up of a beautiful woman, because the AI struggles to separate the core elements from the concept itself.

💡Scene Crafting

Scene crafting is the process of constructing a detailed scene through the use of prompts and text weights, including negative prompting to exclude certain elements. The video demonstrates how to use a combination of positive and negative text weights to create a specific scene, such as a winter landscape with a buffalo herd but without trees or snow.

💡Textual Prompt

A textual prompt is the input given to an AI image generation system to guide the creation of an image. It can include descriptions, concepts, and instructions that the AI uses to generate the desired output. The video discusses how to construct effective textual prompts for Midjourney, including the strategic use of negative prompting.

Highlights

Negative prompting is a process of removing elements from an image using a specific parameter.

Midjourney, a tool for image generation, does not support in-painting or image editing like Dali 2 or stable diffusion.

Negative prompting in Midjourney is subtractive, as opposed to the additive process of regular prompts.

The 'no' parameter can be used to remove specific objects or concepts from generated images.

Remix mode in Midjourney allows working off an existing image to make small changes.

Negative prompting is less effective with complex scenes and cannot edit super specific details.

Using the 'no' parameter is equivalent to applying a negative text weight to a concept in the prompt.

Negative prompting can fail if the concept to be removed is a core part of the image's overall concept.

An abstract prompt allows for more successful negative prompting as it is not tied to a singular concept.

Adding complexity to a prompt with negative prompting can lead to more abstract and less realistic images.

Text weights can be used to emphasize or de-emphasize elements within an image prompt.

Crafting scenes using text weights and negative prompting can build amazing and controlled images.

Negative prompting is not a perfect science and requires trial and error to achieve desired results.

The 'no' parameter can be used multiple times in a prompt to remove different elements.

Negative prompting in Midjourney is currently the only way to remove information from an overall image.

The feature of negative prompts in Midjourney may become more advanced with future updates.

Christian Heidorn from Tokenized AI provides tutorials on navigating the world of AI software.