Multi-Character Scene with Midjourney’s Huge Character Consistency Update (--cref)

Mia Meow
17 Mar 202408:30

TLDRMidjourney has introduced a highly anticipated feature for character consistency, allowing users to generate characters with consistent details using a character reference image. The video demonstrates how to use the new --cref function, which focuses on character traits and works best with characters created by Midjourney. The creator shares a workaround for placing multiple characters in a scene and provides tips on noting down character features to maintain consistency. The process involves using the --cref parameter to control character details, adjusting character weight from 100 (all details) to 0 (just the face), and using the 'vary region' feature to edit images for perfection. The video also offers a method to generate two characters in a scene by being more descriptive in the text prompt. The creator concludes by showing the generated images and encouraging viewers to experiment with the tool for better consistency in character creation.

Takeaways

  • 🎉 Midjourney has released a new feature for character consistency that allows the generation of characters with consistent details using a reference image.
  • 🔍 The new `--cref` function focuses on character traits and is most precise with characters originally created by Midjourney, although it can also be used with real people or photos.
  • 🚫 Midjourney warns that using real people or photos as references may result in distortion, as the feature is not designed for them.
  • 👧 The creator found the `--cref` feature particularly useful for stabilizing facial features, but less so for hair and outfit details.
  • 📈 The results are suitable for creating AI influencers or fashion models, but the focus of the video is on animation-style illustrations.
  • 📝 It's recommended to note down the important features of your characters to maintain consistency as you generate more images.
  • 🌟 The character image can be generated from Midjourney or other sources, but the feature works best with Midjourney-generated characters.
  • 🔗 The image URL can be obtained by dragging the image into the prompt box, right-clicking to copy the image address, or opening the image in a browser and copying the link.
  • ✅ The `--cw` parameter can be used to modify character references, with strengths ranging from 100 (all details) to 0 (just the face).
  • 🔍 Lowering the character weight makes the image adhere more to the text prompt and less to the reference character's hair and outfit.
  • 🖌 The `vary region` feature allows for fine-tuning of specific details in the generated images to better match the original character.
  • 👥 To include multiple characters in a scene, the prompt needs to be more descriptive, specifying the appearance and actions of each character.
  • 💻 For further detail editing, one can use Photoshop's generative tool or similar software to refine the generated images.
  • ⌛ Despite the current limitations, the consistency of character details is expected to improve over time with advancements in the technology.

Q & A

  • What is the new feature introduced by Midjourney for character consistency?

    -The new feature introduced by Midjourney for character consistency is the 'cref' function, which allows users to generate characters with consistent details using a character reference image.

  • What are the limitations of the 'cref' function according to Midjourney?

    -The 'cref' function has limitations in that it won't copy exact details like dimples, freckles, or t-shirt logos. It works best with characters made from Midjourney and is not designed for real people or photos, which may be distorted.

  • Why might the 'cref' feature be more useful for stabilizing facial features rather than hair and outfit details?

    -The 'cref' feature might be more useful for stabilizing facial features because, at the current stage, it is more effective at maintaining consistency in facial traits. The stabilization of hair and outfit details may not meet the preferred standard for animation or storybooks.

  • For what type of content creation is the 'cref' feature particularly suitable?

    -The 'cref' feature is particularly suitable for creating AI influencers or AI models for fashion brands, where the focus is on facial features and the overall style rather than specific hair and outfit details.

  • How does one use the 'cref' function in Midjourney?

    -To use the 'cref' function, one can type in a prompt for the desired image, followed by '--cref' and then insert the image URL of the character reference. The function can be further modified using the '--cw' parameter to adjust the character weight from 100 (all details) to 0 (just the face).

  • What is the purpose of using '--cw' parameter with the 'cref' function?

    -The '--cw' parameter is used to modify the character reference strength, allowing users to control the level of detail that is taken from the reference image, ranging from full character details to focusing solely on the face.

  • How can one obtain the image URL needed for the 'cref' function?

    -To obtain the image URL, one can drag the image directly to the prompt box if it's easily accessible, right-click the image to get the image address, or open the image in a browser and copy-paste the link from the address bar.

  • What is the process to edit the generated images to perfection?

    -To edit the generated images to perfection, one can use the 'vary region' feature to upscale the image and select the area for editing. This allows for adjustments to clothing details, eye gaze, or any other areas that need refinement.

  • How can multiple characters be included in the same scene using the 'cref' function?

    -To include multiple characters in the same scene, one must be more descriptive in the text prompt, specifying the details of each character and their actions. Additionally, the character reference should be switched for each character to be included in the scene.

  • What is the recommended approach when generating images for animation style illustration?

    -When generating images for animation style illustration, it is recommended to focus on the cref function, be descriptive in the text prompt, and use the vary region feature to refine details. The default character weight is usually sufficient unless specific adjustments are desired.

  • How can one ensure consistency in character details when generating multiple images?

    -To ensure consistency in character details, one should note down the important features of the characters and use the cref link with simple prompt descriptions. Additionally, using the vary region feature can help fine-tune details to match the original character ideas.

  • What is the current state of character consistency in Midjourney's generated images?

    -While the character consistency in Midjourney's generated images is not perfect, the use of the cref function significantly improves the consistency compared to using just a reference image. The tool is expected to improve over time, offering better consistency.

Outlines

00:00

🎨 Introducing Character Consistency with Midjourney's 'cref' Feature

The video introduces a new feature by Midjourney that allows for the generation of characters with consistent details using a character reference image. The narrator shares their experience with the feature, noting that while it doesn't perfectly replicate every detail like dimples or logos, it is effective for stabilizing facial features. They find it particularly useful for creating AI influencers and fashion models, and the video focuses on how to use the feature for animation style illustrations. The process involves using the 'cref' function in conjunction with a text prompt for style, and the narrator demonstrates how to generate images using Discord and the Midjourney Alpha website. They also explain how to modify character references using the '--cw' parameter to adjust the level of detail from the reference image.

05:01

🖌️ Refining Character Details and Generating Multiple Characters in a Scene

The second paragraph delves into refining the generated images to better match the desired character details. The narrator discusses using the 'vary region' feature to edit specific parts of the image, such as changing the color of the suspenders on a dress. They then demonstrate how to add a second character to the scene by adjusting the text prompt and switching the character reference to match the new character. The process requires a more descriptive prompt to ensure both characters are included in the generated image. The narrator also suggests using Photoshop's generative tool for further detail editing if needed. The video concludes with the narrator sharing their generated images and inviting feedback on the consistency of the characters, as well as promoting another video with additional character consistency hacks.

Mindmap

Keywords

💡Midjourney

Midjourney is a term used in the script to refer to a specific AI image generation platform. It is the central focus of the video, as the host discusses the platform's new feature for character consistency. This feature allows users to create images of characters with consistent details, which is crucial for animation and storytelling where character recognition is key.

💡Character Consistency

Character consistency refers to the ability to generate images of a character that maintain the same visual traits across different instances. In the context of the video, this is a newly released feature on Midjourney that allows for the creation of characters with stable and recognizable features, which is essential for creating a believable and immersive narrative in animations or storybooks.

💡Character Reference Image

A character reference image is a specific type of image used to guide the AI in generating new images of a character with consistent attributes. In the video, the host uses an image created from Midjourney as a reference to ensure that the generated characters match the desired traits, such as facial features, hair, and clothing.

💡Cref Function

The cref function, as mentioned in the script, is a command used within Midjourney to focus on character traits when generating new images. It is a crucial tool for ensuring character consistency. The host demonstrates how to use the cref function with an image URL to generate images that closely resemble the character reference.

💡Text Prompt

A text prompt is a descriptive input provided to the AI to guide the generation of an image. In the video, the host uses text prompts to describe the scene and actions of the characters, such as 'a little girl picking flowers in the forest in Pixar Animation style,' which helps the AI to create images that match the desired context.

💡Image URL

An image URL is the web address of a specific image that can be used to access or reference that image online. In the context of the video, the host instructs how to obtain an image URL to use with the cref function, which is essential for the AI to reference the character details when generating new images.

💡Character Weight

Character weight is a parameter in Midjourney that allows users to modify the strength of the character reference when generating images. A weight of 100 uses all character details as a reference, while a weight of 0 focuses only on the face. This feature is demonstrated in the video to show how it affects the consistency of character details in the generated images.

💡Vary Region

Vary region is a feature within Midjourney that allows users to edit specific areas of a generated image to refine the details. The host uses this feature to change elements such as the color of the suspenders on a dress to match the original character reference more closely.

💡Upscale Image

Upscaling an image refers to the process of increasing the size of an image without losing quality. In the video, the host upscales images to improve their resolution and then uses the vary region feature to make detailed edits, ensuring that the final images are of high quality and closely match the desired character design.

💡AI Influencers

AI influencers are virtual characters created using AI technology, often used for marketing and social media purposes. The script mentions that the character consistency feature is great for creating AI influencers or AI models for fashion brands, indicating the broader applications of such technology beyond just storytelling and animation.

💡Pixar Animation Style

Pixar Animation Style refers to the distinct visual art style associated with Pixar Animation Studios, known for its high-quality 3D animated films. The host uses this style as an example of the type of animation style that can be emulated using Midjourney's character consistency feature, showcasing how the AI can generate images in various artistic styles.

Highlights

Midjourney has released a new feature focusing on character consistency.

Users can now generate characters with consistent details using a character reference image.

The precision of this technique is limited and does not replicate exact details like dimples or t-shirt logos.

The feature works best with characters created by Midjourney and is not designed for real people or photos.

The creator found the feature more useful for stabilizing facial features rather than hair and outfit details.

The results are suitable for creating AI influencers or fashion brand models.

The video focuses on animation style illustration, which is the preferred style of most viewers.

The cref function in Midjourney allows for the use of character reference images for generating consistent characters.

Noting down important character features helps maintain consistency in generated images.

The character 'Lily' is introduced as an example, with specific features like big round brown eyes and long wavy hair.

The process of generating images using the cref function on Discord is demonstrated.

The --cref parameter is used to insert the image URL of the character reference.

The --cw parameter can modify character references, with strengths ranging from 100 to 0.

Lower character weight focuses more on the text prompt and less on the reference character's details.

Generated images include animals and butterflies from the original image, showing some level of detail consistency.

The 'vary region' feature allows for fine-tuning specific areas of the generated image.

To add a second character to the scene, the prompt must be more descriptive about the characters' appearances.

Switching the character reference to the second character allows for generating a scene with both characters.

Details can be further refined using the 'vary region' feature or Photoshop's generative tool.

The video concludes with a discussion on the current state of character consistency in AI and anticipation for future improvements.