TIPS You NEED for Consistent Characters in Midjourney

Future Tech Pilot
18 Mar 202415:54

TLDRIn this informative video, the host, Nolan, shares valuable tips on creating consistent characters in Midjourney, a tool for AI art creation. He suggests using multiple images as character references to avoid repetitive looks and demonstrates how to manipulate parameters like stylize, chaos, and character weight (CW) to achieve diverse expressions. Nolan also introduces a method to generate multiple expressions by editing images in Photoshop or Canva and uploading them to Discord for character shortcuts. He highlights the importance of not combining character references in a single prompt to avoid blending characters. Additionally, he explores the use of color and texture as character references, showing how Midjourney can integrate these elements into the artwork. The video concludes with advanced tips on combining character and style references and the use of various parameters for unique results. Nolan also mentions his beginner course for mastering Midjourney features.

Takeaways

  • 📸 Use multiple images as character references to avoid the same look in each generation.
  • 🔗 Include multiple image links in the prompt to generate different looks before using them as references.
  • 📈 Experiment with parameters like stylize and chaos values to achieve desired expressions and styles.
  • 👤 Lower the character weight (CW) to focus on the face when developing multiple expressions.
  • 🖼️ Use an image editor to isolate individual faces and save them as separate pictures for character references.
  • 🔄 Create shortcuts for characters by uploading and setting references in Discord, avoiding 'CF' in shortcuts to allow for multiple references.
  • 🎨 Specify new looks directly in the prompt to achieve different character perspectives and expressions.
  • 🚫 Avoid generating multiple characters in a single scene as the AI will combine them; instead, use the 'vary region' feature for inpaintings.
  • 🎭 You can use colors or textures as character references, with the AI smartly incorporating them into the scene.
  • 📉 Be cautious when lowering the character weight as it may result in losing the color intended as a character reference.
  • 🛠️ Combine character and style references with parameters like stylize, chaos, and weird for unique and consistent character images.

Q & A

  • What is the main issue discussed in the video regarding character consistency in mid-journey?

    -The main issue discussed is the frustration of having a character with the same look in each generation, and the video provides solutions to achieve more consistent character appearances.

  • How can multiple image links be used to improve character consistency?

    -By including multiple image links in the prompt, the AI can generate different looks for the character, thus avoiding the repetition of the same appearance.

  • What is the purpose of lowering the stylize value in the prompt?

    -Lowering the stylize value can help in generating more natural and less stylized character expressions, which can contribute to a more consistent character appearance.

  • What is the recommended character weight (CW) when developing multiple expressions for a character?

    -The recommended character weight is to lower it all the way to zero, which allows the AI to focus solely on the face and generate a variety of expressions.

  • How many expressions are suggested to generate for a character?

    -It is suggested to generate between four and six expressions for a character to achieve a good variety without overwhelming the AI.

  • What is the process of isolating individual faces from the generated images?

    -The individual faces need to be isolated using an image editor like Photoshop or Canva, saved as separate pictures, and then uploaded to Discord for character reference.

  • Why is it not recommended to include 'CF' in the shortcut for character references?

    -Including 'CF' in the shortcut limits the use of the image links to character references only and prevents the addition of more than one character reference in the prompt.

  • How can specifying a new look in the prompt affect the character's appearance?

    -By specifying a new look, such as a side profile or portrait, in the prompt, the AI can generate images that match the desired look, providing a more varied and consistent character appearance.

  • Is it possible to generate multiple characters in a single scene using character references?

    -No, generating multiple characters in a single scene using character references directly does not work as the AI will create a combination of both characters. However, a workaround involves using the 'vary region' feature to inpaint the desired characters into the scene.

  • What is the trick mentioned for using color as a character reference?

    -The trick is to set a color or texture as a character reference, and the AI is capable of inserting those colors or textures into the scene, providing a unique and consistent character appearance.

  • How can the 'weird' parameter be used to enhance character consistency?

    -The 'weird' parameter, when combined with other parameters like 'stylize' and 'chaos', can produce unique and consistent character images by introducing an element of unpredictability in the generation process.

  • What is the importance of not including character reference parameters in shortcuts?

    -Including character reference parameters in shortcuts can prevent the AI from recognizing multiple character references in the prompt, limiting the ability to generate images with multiple distinct characters or elements.

Outlines

00:00

🎨 Enhancing Character Consistency with Multiple References

The video begins by addressing the common issue of characters maintaining a similar appearance across different generations. To solve this, the speaker suggests using multiple images as character references, which involves generating various expressions and angles of the character. The process includes adjusting parameters such as stylization and chaos values to achieve desired results. The speaker also recommends lowering the character weight to zero when generating expressions to focus solely on the face. After creating multiple expression images, they should be edited in a program like Photoshop to isolate faces and then uploaded to Discord to create a new character shortcut. The speaker also discusses the importance of not including 'CF' in shortcuts to avoid limitations in character references.

05:00

🚫 Limitations of Multi-Character Scenes and Workarounds

The speaker clarifies that generating multiple characters in a single scene using traditional methods is not possible as the AI will combine them into a single character. However, a workaround is introduced using the 'vary region' feature to inpaint specific characters into a scene. This method involves selecting the desired character's face and inputting their reference links to generate a scene with multiple distinct characters. The video also introduces the viewer to the instructor, Nolan, who offers a beginner course on mastering various features of mid-journey, an AI tool.

10:02

🎨 Creative Use of Color and Texture as Character References

The video explores the innovative use of color and texture as character references in prompts. By including a color or texture link in the prompt, the AI can incorporate those elements into the generated image. The speaker demonstrates this by generating images with characters in armor of specific colors and textures. The importance of not including 'CF' in shortcuts when using color references is emphasized to avoid overriding the character reference with the color. The video also provides additional tips on combining character references with style references and adjusting style weights for better results.

15:04

🔍 Advanced Techniques for Unique Character Generation

The speaker shares advanced techniques for generating unique and consistent character images. These include using a combination of parameters like stylize, chaos, and the 'weird' parameter to create distinctive images. The video showcases examples of these techniques applied to generate impressive character images. The speaker also provides troubleshooting tips for when the AI does not produce the desired results and encourages viewers to explore and find their own tricks. The video concludes with a reminder about the mid-journey beginner course offered by the speaker.

Mindmap

Keywords

💡Character Reference

A character reference in the context of the video refers to using multiple images of a character to guide the AI in generating consistent character appearances across different outputs. This technique helps to overcome the issue of the AI producing the same look for a character in each generation. For example, the script mentions using candid photos and different expressions of the character 'Dash CF' as character references.

💡Stylize Value

The stylize value is a parameter that can be adjusted in the AI's settings to control the level of stylization in the generated images. Lowering the stylize value can lead to more natural-looking results, while increasing it can produce more abstract or artistic outputs. In the script, it is suggested to lower the stylize value to 'S60' for more consistent character expressions.

💡Chaos Value

The chaos value is another parameter that influences the randomness or variability in the AI's generated images. A higher chaos value introduces more unpredictability, which can be useful for exploring a wider range of character expressions and looks. The script provides an example where 'chaos 6' was used to generate diverse results.

💡Character Weight (CW)

Character weight is a parameter that determines the influence of the character reference on the final image. By lowering the character weight to zero, the AI focuses more on generating the face with multiple expressions, which is useful for developing a more nuanced character representation. The script illustrates this with the results of image number three where 'stylize 200' was used.

💡Image Editor

An image editor is a software application used to manipulate or edit images. In the context of the video, an image editor like Photoshop or Canva is used to isolate individual faces from the generated images and save them as separate pictures for further use as character references. This step is crucial for creating a shortcut for the character in the AI system.

💡Discord

Discord is a communication platform that the video script mentions as a place to upload the edited images for character references. It is used as a medium to create and manage shortcuts for character references, which can be included in the AI's prompts to generate images.

💡Shortcut

In the video, a shortcut refers to a predefined set of parameters or image links that can be quickly inserted into the AI's prompt to generate images. Creating a shortcut for a character, such as 'Eliza 2', simplifies the process of including multiple character references in the AI's prompts.

💡Vary Region

Vary region is a feature that allows users to specify different regions within an image for the AI to focus on during the generation process. This technique is used to inpaint or insert multiple characters into a scene by selecting their faces and changing the character references accordingly.

💡Color as Character Reference

The video introduces an innovative technique where color can be used as a character reference. The AI is capable of understanding the color reference and incorporating it into the scene, as demonstrated by the examples where 'sky blue' and 'turquoise' colors were used to influence the appearance of the characters' armor.

💡Style Weight (SW)

Style weight is a parameter that controls the influence of a style reference image on the final output. Lowering the style weight, as suggested in the script with 'D-S SW20', can help blend the style with the character reference more subtly, leading to unique and consistent character images.

💡Weird Parameter (D-W)

The weird parameter is an experimental setting in the AI that can produce unexpected and unique results by adjusting the level of 'weirdness' in the generated images. The script provides an example where 'weird 340' was used in combination with other parameters to create a striking character image.

Highlights

Use multiple image links as character references to create consistent characters.

Generate different looks by including candid photos and varying expressions.

Adjust the stylize value and chaos parameters for diverse character results.

Lower the character weight (CW) to zero when developing multiple expressions.

Isolate individual faces using an image editor and upload them separately.

Create a new shortcut for the character with all solid reference images.

Avoid using 'D-CF' in shortcuts to prevent limitations on character references.

Specify new looks in the prompt for different character expressions.

Upscale preferred images and include them in the character shortcut.

Generating multiple characters in a scene is not possible; the bot combines them.

Use 'vary region' and 'inpaint' features to include multiple characters.

Color can be used as a character reference, influencing the scene's color scheme.

Set colors as shortcuts for easier use in prompts.

Combining character reference with style reference can yield unique results.

Experiment with parameters like stylize, chaos, and weird for varied outputs.

Lowering the character weight may result in losing the color reference.

Troubleshooting tips are available for when Midjourney isn't providing desired results.

Nolan's mid-journey beginner course offers detailed lessons on every feature.