Underused Midjourney v5 Prompt Commands :: How to use Text Weight and Image Weight

Theoretically Media
29 Mar 202311:38

TLDRIn this video, the presenter discusses the underutilized yet powerful techniques of Image Weight and Text Weight in mid-journey prompts. The video explains how the mid-journey system assigns tokens to keywords in prompts and emphasizes the importance of the order of keywords. Text weights are introduced as a method to control the emphasis on specific keywords by using a colon colon followed by a number to indicate the weight. The video also covers the correct format for text weights and provides examples of how they can be used to influence the composition of generated images. Image weights are then explored, explaining how they can be used with reference images to guide the style and composition of the output. The presenter shares examples of using image weights with different values and discusses the limitations and workarounds when using negative prompts. The video concludes with a demonstration of photobashing to achieve desired results when direct text prompts fail.

Takeaways

  • πŸ“ **Understanding Weights**: Text and Image Weights are powerful techniques in mid-journey prompts to control the output by assigning more tokens to specific keywords or reference images.
  • πŸ” **Prompt Scanning**: Mid-journey scans prompts for keywords and assigns tokens, placing more emphasis on words at the beginning of the prompt.
  • βš–οΈ **Text Weight Syntax**: Use '::' followed by a number to weight keywords, with no space between the keyword and '::', and a space after the number.
  • πŸ”’ **Text Weight Examples**: Weighting the first keyword as '::1' and the second as '::2' makes the second word twice as important.
  • 🍰 **Example of Text Weight**: The transcript provides an example of how the word 'cupcake' can be weighted to influence the composition of the generated image.
  • πŸ–ΌοΈ **Image Weight Usage**: Reference images can be weighted using '--IW' followed by a number between 0.5 and 2 to indicate reliance on the reference image.
  • 🎨 **Image Weight Impact**: Adjusting image weight can significantly change the style and composition of the generated image, as demonstrated with the Scarlett Johansson example.
  • πŸ”„ **Negative Prompting**: Negative prompts with '::-' can be used to remove unwanted elements from the generated image, although results may vary.
  • 🧩 **Photobashing**: When direct text prompts fail, photobashing techniques can be used to manually adjust the generated image to fit the desired outcome.
  • πŸš€ **Compositional Balance**: Weights allow for fine-tuning the balance between elements in a complex prompt, as shown with the 'Samurai' example.
  • πŸ“ˆ **Iterative Process**: Achieving the desired image may require several attempts and adjustments to both text and image weights.
  • πŸ“š **Learning from Examples**: The video provides practical examples and encourages viewers to experiment with different weights for better results.

Q & A

  • What are the two techniques discussed in the video that can be used to control the output in mid-journey prompts?

    -The two techniques discussed in the video are Image Weight and Text Weight, which are used to control the emphasis on different elements within a prompt to influence the generated image.

  • How does the mid-journey system assign tokens to keywords in a prompt?

    -Mid-journey scans the prompt for keywords and assigns tokens to those keywords. It uses these tokens to assemble the image based on its database knowledge. The system places more emphasis on words at the beginning of the prompt than those at the end.

  • What is the maximum number of tokens typically assigned per prompt in mid-journey?

    -The speaker mentions they've heard that about 75 tokens are assigned per prompt, but this number may have increased as the mid-journey team refines the language model.

  • How can text weights be applied in a mid-journey prompt?

    -Text weights are applied by using a double colon (::) instead of a comma to separate keywords, followed by a number that represents the weight. For example, 'keyword1::2, keyword2::1' would make 'keyword1' twice as important as 'keyword2'.

  • What is the recommended range for text weights when using them in a prompt?

    -The speaker recommends using numbers between one and ten for text weights for simplicity, although technically any number or even a decimal point can be used.

  • How does the format of text weights affect the mid-journey prompt?

    -The format requires no space between the keyword and the double colon, but there must be a space after the weighted number. If formatted incorrectly, mid-journey may ignore the weight.

  • How does the use of image weights in mid-journey differ from stable diffusion's posed image?

    -Image referencing in mid-journey is not a direct one-to-one replication like stable diffusion's posed image. Instead, the output is inspired by the reference image, creating a result that is influenced by the image weight assigned.

  • What is the purpose of using an image reference in mid-journey prompts?

    -An image reference serves as a guide for the style or composition of the generated image. It is used by placing the image URL at the front of the prompt and applying an image weight to indicate the level of reliance on the reference image.

  • How can negative prompts be used to modify the output in mid-journey?

    -Negative prompts are used by adding a keyword followed by a double colon and a negative number. This tells mid-journey to avoid including those elements in the generated image.

  • What is the main challenge when trying to remove certain elements using negative prompts?

    -The challenge is that mid-journey might not fully comply with the negative prompt, especially if the element to be removed is present in the image reference or is strongly associated with the prompt's context.

  • What alternative method was suggested to overcome the limitations of negative prompting?

    -The speaker suggested using photobashing, which involves manually editing an image to remove or alter specific elements, as an alternative method to achieve the desired outcome when negative prompting is not effective.

  • What is the importance of experimenting with different weights and prompts in mid-journey?

    -Experimenting with weights and prompts allows users to fine-tune the composition and style of the generated images, helping to achieve more accurate and desired results.

Outlines

00:00

πŸ˜€ Understanding Weights in Mid-Journey Prompting

This paragraph introduces the concept of 'Image Weight' and 'Text Weight' in the context of mid-journey prompting. It explains how these weights can be used to control the output of an image generation system by assigning more tokens to specific keywords. The video aims to clear up confusion around these techniques, which are often underutilized. It also discusses how the system prioritizes keywords at the beginning of a prompt over those at the end and how weights can be applied using a colon colon syntax followed by a number. The importance of correct formatting for weights is emphasized, and examples are provided to illustrate the impact of weights on the generated images.

05:01

🎨 Manipulating Composition with Weights and Reference Images

The second paragraph delves into the practical application of weights to manipulate the composition of generated images. It discusses how increasing the weight of certain elements, like 'katana' or 'mystical forest', can shift the focus of the image. The use of reference images in mid-journey is introduced, where an image URL is added to the prompt to influence the output. The paragraph provides examples of how adjusting the image weight (IW) can lead to different styles and compositions in the final image. It also touches on the limitations of negative prompting and the workaround of photobashing to achieve desired results.

10:01

πŸ” Experimenting with Negative Prompts and Photobashing

The final paragraph explores the use of negative prompts to exclude certain elements from the generated images. It humorously recounts the difficulty of removing a hat from an image, despite the presence of a hat in the reference image. The paragraph demonstrates the limitations of text prompts in overriding the influence of reference images and concludes with a successful photobash of Clint Eastwood's face onto a Harrison Ford body. This section encourages experimentation and provides a workaround for when the AI does not respond as expected to negative prompts.

Mindmap

Keywords

πŸ’‘Midjourney

Midjourney refers to an AI image generation tool that uses prompts to create images. In the video, it is the primary subject where the host discusses how to effectively use it to control the output of generated images through text and image weights.

πŸ’‘Text Weight

Text weight is a technique used in Midjourney prompts to assign different levels of importance to keywords within the prompt. By using a colon followed by a number, users can tell the AI to focus more on certain elements. It is a crucial aspect of controlling the composition of the generated images.

πŸ’‘Image Weight

Image weight is another technique used in conjunction with reference images in Midjourney. It allows users to control the influence of the reference image on the final output by specifying a weight between 0.5 and 2. This helps in achieving a balance between the reference image and the textual prompt.

πŸ’‘Prompts

Prompts are the textual inputs given to the Midjourney AI to generate images. They are composed of keywords that the AI interprets and uses to create images. The video discusses how the order and weight of these keywords within the prompt affect the final image.

πŸ’‘Tokens

In the context of Midjourney, tokens are units assigned to keywords in a prompt. The AI uses these tokens to understand the importance of each keyword and assemble the image accordingly. The script mentions that about 75 tokens are assigned per prompt.

πŸ’‘Compositional Balance

Compositional balance refers to the visual arrangement or layout of elements within an image. The video explains how text and image weights can be used to adjust this balance, ensuring that the generated images align with the user's creative vision.

πŸ’‘Reference Image

A reference image is an existing image that a user can upload and use as a guide for the AI to generate a new image. In the video, the host demonstrates how to use reference images with image weights to inspire the style or elements of the generated image.

πŸ’‘Negative Prompt

A negative prompt is a technique where users specify elements they do not want to appear in the generated image by using a negative number with the text weight. The video shows examples of how this can be tricky, as the AI may still include unwanted elements due to the influence of the reference image.

πŸ’‘Photobashing

Photobashing is a manual editing technique where users combine or edit parts of different images to achieve a desired result. In the video, the host uses photobashing to remove an unwanted hat from an image generated by Midjourney.

πŸ’‘Illustrator Style

The term 'illustrator style' in the video refers to a specific visual aesthetic often associated with comic books or graphic novels. The host discusses using image weights to incorporate elements of a particular illustrator's style, such as Jaylee's, into the generated images.

πŸ’‘Dynamics

Dynamics in the context of the video refers to the energy and movement conveyed in the generated images. The host experiments with image and text weights to achieve more dynamic poses and compositions in the images created by Midjourney.

Highlights

Introduction to two powerful techniques in mid-journey prompting: Image Weight and Text Weight.

Explanation of how prompts work in mid-journey by scanning for keywords and assigning tokens.

Tokens are not assigned equally; more emphasis is placed on words at the start of the prompt.

Text weights allow users to add more tokens to specific keywords by using a colon colon followed by a number.

Image of a cupcake is used to demonstrate the effect of text weights on the assembled image.

The importance of correct formatting when using text weights to avoid being ignored by mid-journey.

Demonstration of how image weights can affect the compositional balance in a longer prompt.

Use of an old Samurai example to show the impact of different weights on the final image.

Mention of the limitations of text weights and strategies to work around them.

How to use reference images in mid-journey by uploading an image and using it as a reference in the prompt.

Exploring image weights with a Scarlett Johansson example and adjusting the weights to achieve different styles.

The difference between image referencing and stable diffusion's posed image in mid-journey.

A project example involving a fictional documentary about 'The Dark Tower' movie to illustrate the use of image references.

Experimenting with negative prompts to remove specific elements from the generated image.

Challenges faced when trying to remove a hat from an image using negative prompts and the workaround using photobashing.

The presenter's recommendation to use natural language for simpler and more effective prompts.

A call to action for viewers to like, subscribe, and support the channel for more content.

Final thoughts and thanks from the presenter, Tim, for watching the video.