How to Write the Most Accurate Midjourney Prompts - /describe Beginner Tutorial

Future Tech Pilot
9 Jun 202307:16

TLDRThe video tutorial explains how to write accurate prompts for generating images using the 'describe' feature in Discord's Midjourney. It starts by suggesting the direct use of the feature to obtain initial prompts based on an image. The process involves uploading the image to Discord and using the generated options to submit prompts. If the results aren't satisfactory, the video recommends adding the original image as a reference and adjusting the weight of the reference image in the prompt to improve the accuracy of the generated images. It also covers adjusting the stylize and chaos values for more literal or diverse outputs, respectively. The tutorial ends with a note on the limitations of the 'describe' feature, such as its inability to recreate text from images, and the creator's offer of a prompt pack for purchase.

Takeaways

  • 🖼️ To write a prompt for an image, use the 'describe' feature in Discord with Midjourney by uploading the image and selecting from the given options.
  • 🔍 If the initial results are unsatisfactory, add the original reference picture as an image prompt to improve the accuracy of the generated images.
  • 📎 Copy the image address and paste it into the prompt to use the reference picture as a foundational picture for the prompts.
  • 🔗 Increase the weight of the reference picture by using '--IW' followed by a number between 0.5 and 2 to control its importance in the image generation process.
  • 🎨 Adjust the 'stylize' value to make the generated images more literal or creative, and use the 'chaos' value to introduce variety in the generated grid.
  • 🔄 If you're not satisfied, use the 're-roll' feature on the describe prompt to get four new variations.
  • 💡 The describe feature uses a model not native to Midjourney, so it may not perfectly recreate images even if it describes them accurately.
  • 📈 Experiment with different stylize and chaos values to find the right balance between adherence to the prompt and creative freedom.
  • 📉 A lower stylized value results in more literal images, which can be less aesthetically pleasing if the prompt is not well-crafted.
  • 🌐 Be aware that aspect ratios may not be exact due to Midjourney rounding to the nearest 32-pixel value.
  • 🎭 The describe feature can interpret made-up names, suggesting it has some capacity for understanding context beyond existing data.
  • 📦 The creator offers a prompt pack with 51 favorite prompts and 69 examples for a fee, aiming to save time and demonstrate how to create impressive images.

Q & A

  • What is the first step to use the describe feature in Discord with an image?

    -The first step is to save the image on your computer, then use the 'describe' feature by typing 'forward slash describe' in Discord, and upload the image when prompted.

  • How many options does the describe feature typically provide after uploading an image?

    -The describe feature typically provides four different options after uploading an image.

  • What can be done if the initial results from the describe feature do not resemble the original picture enough?

    -If the initial results do not resemble the original picture, you can add the original reference picture as an image prompt, adjust the weight of the reference picture, or modify the stylize and chaos values to refine the results.

  • How do you add the original reference picture to the describe prompts?

    -To add the original reference picture, click the plus button next to the prompt box, upload the image, then right-click on the uploaded image and copy its address. Paste this link into the describe prompts when submitting.

  • What is the purpose of adding a weight to the reference picture?

    -Adding a weight to the reference picture allows you to control the importance of the reference image in the generation process. A higher weight means the reference picture will have a greater influence on the output.

  • What is the range for the weight that can be assigned to the reference picture?

    -The weight can be a number between 0.5 and 2.0, where 0.5 makes the reference picture less important and 2 makes it twice as important as the other prompts.

  • How can adjusting the stylize value affect the output of the describe feature?

    -Lowering the stylize value makes the output more literal and closely follows the prompt, while raising it gives more creative freedom to the system, resulting in more artistic and less predictable outputs.

  • What is the default chaos value and how does it affect the describe feature's output?

    -The default chaos value is zero, which results in outputs that are more similar to each other. Increasing the chaos value introduces more variety among the generated images.

  • What does the 're-roll' option in the describe feature allow you to do?

    -The 're-roll' option allows you to generate four new prompts that are similar but not identical to the previous ones, which can help in refining a particular style.

  • Why might the describe feature sometimes mention an artist with a hyperlink and sometimes without?

    -The describe feature might mention an artist with a hyperlink if the artist is real and has a known online presence. If there is no hyperlink, it could be a made-up name that the system has learned to interpret.

  • How does the aspect ratio in the describe feature work and why might it not match the exact ratio you input?

    -The aspect ratios in the describe feature are rounded to the nearest 32-pixel value. Therefore, an input like 16 by 9 might not result in an exact ratio due to this rounding.

  • What is the limitation of the describe feature when it comes to generating text within an image?

    -The describe feature can recognize and describe text within an image, but it cannot accurately recreate that text in the generated output, even if it understands what the text is supposed to be.

Outlines

00:00

🖼️ Image Prompt Creation with Discord's Describe Feature

The first paragraph introduces a method for creating prompts for images using Discord's 'describe' feature. It explains how to ask for a description of an image saved on your computer and choose from four generated options. The user is then guided on how to refine the prompts by adding a reference picture, adjusting the importance of the reference through a weighting system, and modifying the 'stylize' and 'chaos' values for more creative freedom or closer adherence to the prompt. The paragraph also discusses the limitations of the 'describe' feature, such as its inability to recreate text from images.

05:01

🎨 Advanced Techniques for Image Generation with Mid-Journey

The second paragraph delves into advanced techniques for generating images using Mid-Journey's 'describe' feature. It covers the use of the 're-roll' option to get additional prompts, the interpretation of made-up artist names by the 'describe' feature, and the quirks of aspect ratio handling in Mid-Journey. The speaker also clarifies that while 'describe' can accurately describe an image, it may not always generate a picture that perfectly matches the description, particularly when text is involved. The paragraph concludes with a prompt for viewers to like the video to help share it with a wider audience and a sign-off wishing well to the viewers.

Mindmap

Keywords

Midjourney

Midjourney refers to a hypothetical AI model or tool that is capable of interpreting and generating images based on textual prompts. In the context of the video, it is used to describe an AI's ability to analyze an image and provide descriptive prompts for it, which can then be used to generate new images that are more closely related to the original picture.

Discord

Discord is a popular communication platform that supports text, video, and voice conversations. In the video, it is mentioned as the platform where the Midjourney's 'describe' feature is accessed, allowing users to interact with the AI and upload images for analysis.

Describe Feature

The 'describe' feature is a function within the Midjourney tool that analyzes an uploaded image and provides a set of descriptive prompts based on what the AI 'sees' in the image. It is central to the tutorial as it helps users generate more accurate and relevant prompts for image generation.

Image Prompt

An image prompt is a reference image used to guide the AI in generating new images. In the video, adding an original reference picture as an image prompt helps the AI to create images that more closely resemble the original, thus improving the accuracy of the generated content.

Weight

In the context of the video, 'weight' refers to the importance assigned to the reference picture in the image generation process. By adjusting the weight (using --IW followed by a number between 0.5 and 2), users can control how much influence the reference picture has on the final generated images, ensuring a closer match to the desired outcome.

Stylize Value

The stylize value is a parameter that can be adjusted to make the AI follow the prompt more closely (by setting it to zero) or to allow for more creative freedom (by increasing it). It affects how literal or abstract the generated images will be in relation to the prompt provided.

Chaos Value

Chaos value is another parameter that introduces variability into the image generation process. By increasing the chaos value (up to a maximum of 100), the AI produces images with greater diversity, ensuring that each generated image is distinct from the others.

Aspect Ratio

The aspect ratio refers to the proportional relationship between the width and the height of an image. In the video, it is mentioned that Midjourney's outputs round to the nearest 32-pixel value, which may result in slight deviations from the exact aspect ratio specified by the user.

Re-roll

Re-roll is a feature within the Midjourney tool that allows users to generate additional prompts based on the same input, resulting in a new set of image options. This can be useful for refining the style and finding the most suitable outcome.

Prompt Pack

A prompt pack is a collection of pre-made prompts that can be used to generate images with Midjourney or similar AI tools. In the video, the creator offers a prompt pack containing their favorite prompts and examples to save viewers time and provide a starting point for creating high-quality images.

Squire Zed

Squire Zed is mentioned as a contributor on Discord who has provided valuable insights or feedback regarding the use of the 'describe' feature. The mention acknowledges the collaborative nature of the community and the importance of shared knowledge in refining the use of AI tools.

Highlights

Using the 'describe' feature in Discord to ask Midjourney what it sees in an image.

Uploading an image to Discord and receiving four different prompt options.

Adding the original reference picture as an image prompt to improve the accuracy of the generated prompts.

Copying the image address and pasting it into the describe prompts to use as a foundational picture.

Adjusting the weight of the reference picture with the --IW parameter to control its importance in the prompt.

Experimenting with different weights to see how it affects the direction and alignment of the generated images.

Adjusting the stylize value to make Midjourney follow the prompt more closely or allow more creative freedom.

Adding chaos value to introduce more variety in the generated grid of images.

The impact of setting the stylized value to zero and its effect on the literalness of the generated images.

The difference in appearance when the chaos value is increased to 40 and 100.

Announcement of a prompt pack available for purchase containing 51 favorite prompts with 69 examples.

The option to re-roll on the describe feature for additional prompts.

Mention of artist names in the describe prompt, with or without hyperlinks, indicating Midjourney's ability to interpret made-up names.

Explanation of aspect ratios in Midjourney outputs and how they round to the nearest 32-pixel value.

The limitations of Midjourney's describe feature in accurately recreating images, even when it can describe them perfectly.

The importance of keeping in mind that describe uses a model not inherently native to Midjourney for image description.

A call to action for viewers to like the video to share it with more people interested in Midjourney's describe feature.