HUGE Midjourney V6 Update | NEW Character Ref (--cref), Style Ref (--sref), & Image-to-Text Updates!

Woollyfern
26 Mar 202408:25

TLDRMidjourney V6 has introduced significant updates, including four versions of Style Reference (SRF) for applying artistic styles to generated images. The latest default, SRF version 4, offers improved performance and style accuracy. The Image-to-Text model, 'describe', has been optimized for V6, providing text prompts based on input images to inspire new ideas. Additionally, the new Character Reference feature aims to maintain character consistency across multiple image generations, although it may require prompt adjustments for best results. These updates enhance creative possibilities for storytelling and artistic expression within the Midjourney platform.

Takeaways

  • 🎨 Midjourney has introduced four versions of Style Reference (SRF), which allows users to apply the artistic style of a reference image to their results.
  • 🔄 SRF versions have evolved to improve performance and reduce the inadvertent inclusion of unwanted elements from the reference image.
  • 📈 The newer SRF versions provide a more linear performance at lower style weight values, offering better control over the style application.
  • 🖼️ The Image-to-Text model, 'describe', has been optimized for Midjourney's V6 model to generate text descriptions that better align with the model's capabilities.
  • 📝 'Describe' can be a useful tool for generating prompts or gaining inspiration based on the interpretation of an image.
  • 🧑 The Character Reference feature is designed to maintain character similarity across multiple image generations, which is beneficial for visual storytelling.
  • 🔍 Character Reference works best with images made by Midjourney and characters with notably distinct features, allowing for creative exploration of character portrayal.
  • 📌 The Character Weight parameter controls the strength of the character reference, with options to lock onto facial features or adjust clothing styles.
  • 🛠️ Character Reference is not perfect and may require workarounds through prompting to achieve desired results, such as avoiding unnatural angles.
  • 📈 The first version of Character Reference is expected to improve over time, offering more accurate character replication in future updates.
  • ✅ The video provides a comprehensive overview of the latest Midjourney updates, encouraging viewers to experiment with the new features and share their experiences.

Q & A

  • What are the four versions of style reference mentioned in the video?

    -The four versions of style reference are: version one, which was the original and default until March 14th; version two, which became the default on March 14th and lasted until March 21st; version three, an upgraded release of version one for better performance; and version four, an upgraded release of version two, which is currently the default.

  • How can you specify which version of style reference to use in Midjourney?

    -You can specify which version of style reference to use by including the new parameter `--SV` in your prompt. If you don't specify a version, the current default will be used.

  • What are the two main issues the updates aim to address with style reference?

    -The updates aim to address improved performance to prevent inadvertently bringing in unwanted elements from the reference image, and refinement of the style weight parameter to provide a more linear performance at lower style weight values.

  • What is the 'describe' feature in Midjourney and how can it be used?

    -The 'describe' feature in Midjourney is an image-to-text model that generates text descriptions of an image. It can be used to find the right words to describe an image or for inspiration for prompts. To use it, type `/describe` followed by the image URL or upload an image file, and Midjourney will generate four text descriptions.

  • How does the character reference feature in Midjourney work?

    -The character reference feature in Midjourney is used to maintain character similarity across multiple image generations. You include the `--CF` parameter at the end of your prompt along with a URL to the character you want to reference. Midjourney will then attempt to replicate that character in the results.

  • What is the range of the character weight parameter and how does it affect the character reference?

    -The character weight parameter ranges from 0 to 100 with a default value of 100. When set to 100, Midjourney locks onto the face and clothing of the reference character. When set to 0, it only locks onto the face, allowing changes to clothing while retaining the character's facial features.

  • How can you optimize the character reference feature to avoid issues like unnatural angles or incorrect details?

    -You can optimize the character reference feature by carefully wording your prompt to avoid pushing Midjourney towards creating images with undesired angles or details. For example, changing 'walking through' to 'standing' and specifying details like 'looking at the camera' or 'wearing a red t-shirt' can help achieve better results.

  • Can the character reference feature be used with non-human characters?

    -Yes, the character reference feature can be used with non-human characters as well. The feature focuses on capturing the overall visual characteristics of the character, regardless of whether the character is human or not.

  • What is the purpose of the 'describe' feature in generating prompts for Midjourney?

    -The 'describe' feature serves to provide starter prompts that give clues about the words or phrases that Midjourney considers most effective in describing an image. These prompts are not exact recreations of the input image but offer insights into powerful descriptors that can be used in creating new prompts.

  • What is the default style reference version currently in use by Midjourney?

    -The default style reference version currently in use by Midjourney is version four, which is an upgraded release of version two.

  • How can you submit all four text descriptions generated by the 'describe' feature at once?

    -If you have fast mode turned on, you can click 'imagine all' to submit all four text descriptions generated by the 'describe' feature at once.

  • What does the character weight parameter control in the character reference feature?

    -The character weight parameter controls the strength of the character reference. It allows for adjustments in how closely the generated image adheres to the clothing and facial features of the reference character.

  • What does the 'reroll' button do in the 'describe' feature?

    -The 'reroll' button in the 'describe' feature allows you to generate a new set of text descriptions for the image, providing different prompts for inspiration or use in Midjourney.

Outlines

00:00

🎨 Mid Journey Style Reference Updates

The video discusses the recent updates to Mid Journey, focusing on the new versions of style reference. There are now four versions available, with version 4 being the current default. Each version offers different levels of performance and style accuracy. The updates aim to improve performance by reducing the inadvertent inclusion of unwanted elements from the reference image and enhancing the style weight parameter for a more linear effect. The viewer's preference is for version 4 due to its accuracy, but version 3 is also appreciated for its creative results.

05:01

📄 Image-to-Text Model 'Describe' Feature

The 'describe' feature in Mid Journey, which converts images to text prompts, is highlighted. Initially designed for the V5 model, it has been optimized for V6 to generate text descriptions that better align with the model's capabilities. The feature is useful for generating inspiration for prompts or understanding how Mid Journey interprets images. By using the 'describe' command, the system provides four text descriptions that can be used as starting points for image prompts. The repeated words or phrases in these descriptions may indicate powerful descriptors for the image.

🧑 Character Reference Feature

The newly introduced character reference feature is explained, which is used to maintain character consistency across multiple image generations. While not perfect, it captures the overall visual characteristics of a provided character reference image. The feature works best with Mid Journey-generated images and characters with distinct features. It can also be used with non-human characters. The character weight parameter allows for control over the strength of the reference, with 100 being the default value that locks onto the face and clothing of the character. The video also discusses workarounds for issues like unnatural character angles by adjusting the text prompt and mentions that the feature is expected to improve over time.

Mindmap

Keywords

Midjourney V6 Update

The Midjourney V6 Update refers to the latest improvements and additions to the Midjourney software, which is likely an AI-driven image generation tool. The update includes new features and enhancements that aim to improve the user experience and the quality of image generation. It is central to the video's theme as it outlines the advancements made to the software.

Character Reference (--cref)

Character Reference, denoted by (--cref), is a new feature that allows users to maintain character similarity across multiple image generations. It is significant for creating visuals for stories where the same character appears in various scenes or outfits. This feature is highlighted in the video as it represents a major update in the software's capabilities.

Style Reference (--sref)

Style Reference, indicated by (--sref), is a feature that enables the application of an artistic style from a reference image to the results generated by Midjourney. It is crucial for controlling the visual aesthetics of the generated images. The video discusses four versions of this feature, showing its evolution and importance in the update.

Image-to-Text Updates

Image-to-Text Updates refer to improvements made to the 'describe' feature of Midjourney, which generates text descriptions based on an input image. This feature is valuable for users who need inspiration for their prompts or are trying to understand how the AI interprets different elements of an image. The update has optimized the feature for better compatibility with the V6 model.

Style Weight Parameter (--s SW)

The Style Weight Parameter, represented by (--s SW), is a setting that controls the intensity of the style reference's impact on the generated image. It is mentioned in the context of how the newer versions of the style reference feature have improved the performance at lower style weight values, making the results more linear.

Describe Feature

The Describe Feature is an image-to-text model within Midjourney that provides text descriptions for an input image. It is used to help users find the right words to describe an image or to inspire prompt creation. The video emphasizes its optimization for the V6 model and its utility in generating starter prompts.

Character Weight Parameter (--CW)

The Character Weight Parameter, denoted by (--CW), is a control that adjusts the strength of the character reference in the generated images. It ranges from 0 to 100, with 100 locking onto both the face and clothing of the reference character, and 0 focusing only on the face. This parameter is essential for fine-tuning the character representation in the output.

Discord Announcement

The Discord Announcement is a communication channel where Midjourney shares updates and information about new features with its users. It is mentioned in the context of listing the four versions of the style reference feature, indicating its role as a source of official updates and community engagement.

Artistic Style

Artistic Style refers to the unique visual elements, such as color, medium, and overall aesthetics, that are characteristic of an artist's work or a specific piece of art. In the context of the video, the artistic style is applied to the generated images through the style reference feature, allowing users to emulate the style of a given reference image.

Performance Improvement

Performance Improvement in the video script refers to the enhancements made to the Midjourney software that result in better image generation quality and fewer unintended elements being included from the reference image. It is a key focus of the V6 update and is discussed in relation to both the style and character reference features.

Fast Mode

Fast Mode is a feature within Midjourney that allows users to quickly submit all four text descriptions generated by the describe feature as image prompts. It is mentioned as a convenient option for users who want to utilize the generated prompts without having to submit them individually.

Highlights

Mid Journey has released a V6 update with new features including Character Ref (--cref), Style Ref (--sref), and improvements to the Image-to-Text model.

Four versions of Style Reference (SRef) are now available, with version 4 being the current default.

The new versions of Style Reference aim to improve performance and reduce unintended elements in the generated images.

The Style Weight parameter (--sSW) has been enhanced for a more linear performance at lower values.

The Image-to-Text model 'describe' has been optimized for the V6 model to better assist in generating prompts.

The 'describe' feature can generate four text descriptions to aid in finding effective descriptors for an image.

Character Reference feature allows maintaining character similarity across multiple image generations.

Character Reference is particularly useful for creating visuals where the same character appears in different scenes.

The Character Weight parameter (--CW) controls the strength of the character reference in the generated images.

Character Reference can be used with both human and non-human characters.

The Character Reference feature is not perfect and may require workarounds through prompting.

The video provides tips for using Character Reference effectively, such as adjusting the text prompt to avoid unnatural angles.

The Character Reference feature is expected to improve with future updates.

The video offers a comprehensive breakdown of the new features and provides guidance on how to use them.

The updates aim to streamline the prompting process and provide more accurate and creative results.

The video invites viewers to share their experiences with the new features and suggest topics for a deep dive video.

The video emphasizes the importance of keeping up with the latest features and updates in Mid Journey.