Midjourney v6.1 vs v6.0: What You NEED to Know | SREF, Beauty Bias, New --p Codes | Midjourney Guide

Woollyfern
8 Aug 202406:18

TLDRMidjourney's v6.1 update enhances image quality with reduced pixel artifacts and more realistic textures, especially in eyes and skin. It introduces a Q2 mode for additional details at the cost of slight coherence loss. The new upscaler shows promise but may not surpass Topaz Labs' performance. While v6.1 exhibits a conventional beauty bias, making characters appear smoother, Midjourney is aware and plans to address it in v7. Users can still utilize SR codes and character references effectively. The model personalization system in v6.1 is more nuanced, adapting to visual preferences based on image rankings. Despite the improved quality, some prefer v6.0 for its creativity and diversity.

Takeaways

  • 🆕 Midjourney has released a new v6.1 model, which is now the default for all users, but can be changed in settings.
  • 🖼️ The v6.1 update enhances image quality, with improvements in texture, clarity, and realism, especially for eyes and skin.
  • 🔍 The coherence and accuracy of body and hand depictions have been improved in v6.1.
  • 📈 The introduction of a new --Q2 mode in v6.1 allows for more detailed textures at the expense of slightly reduced image coherence.
  • 🆚 A comparison of Midjourney's 2X upscaler with Topaz Labs' Gigapixel shows varying results, with Topaz excelling in skin texture and eye clarity.
  • 💆‍♀️ v6.1 exhibits a more pronounced conventional beauty bias, with smoother skin and fewer imperfections, a side effect of using user data to improve image quality.
  • 🎭 Users can still use previous character references with v6.1, but may notice a slight increase in conventional beauty bias.
  • 🎨 Style references (SR codes) should yield the same style in v6.1, but with potentially less detail and complexity compared to v6.0.
  • ✅ Text rendering accuracy is reported to be improved in v6.1, though more user feedback is needed for confirmation.
  • 🔄 The tile parameter for creating seamless patterns has been fixed in v6.1, maintaining pattern integrity even after upscaling.
  • 🤖 The new model personalization system in v6.1 is designed to be more accurate and nuanced, adapting to user preferences based on image rankings.

Q & A

  • What is the main difference between Midjourney v6.1 and v6.0 in terms of image quality?

    -Midjourney v6.1 brings a boost to image quality with reduced pixel artifacts and enhanced textures. Eyes and skin have improved clarity and look more realistic in photographic images. There is also better coherence and accuracy with elements like bodies and hands.

  • What is the purpose of the '--Q2' parameter in Midjourney v6.1?

    -The '--Q2' parameter in Midjourney v6.1 is used to sometimes get more texture and details in the results at the cost of slightly reduced image coherence.

  • How does the upscaler in Midjourney v6.1 compare to Topaz Labs' Gigapixel?

    -In a side-by-side comparison, both Midjourney's upscaler and Topaz Labs' Gigapixel did a decent job. Midjourney was particularly good at fixing eyelashes and hair texture, while Topaz provided better skin texture, sharpness, and clarity of the eyes.

  • Does Midjourney v6.1 have a different approach to beauty bias compared to v6.0?

    -Yes, v6.1 has a bit more of a conventional beauty bias than v6.0, resulting in smoother skin and fewer imperfections. This was an unintended side effect of incorporating user-generated data to boost image quality.

  • Can you use character references from v6.0 in v6.1?

    -Yes, you can use any of your previous character references with v6.1, and it generally handles character references just like v6.0.

  • How does the style rendering differ between v6.0 and v6.1 when using SR codes?

    -The images rendered with SR codes in v6.1 should have the same style as v6.0 because they are working with the same style space. However, v6.1 images might look a little more simplistic and less detailed.

  • What improvements were made to text rendering accuracy in Midjourney v6.1?

    -Text rendering accuracy should be improved in v6.1, although the speaker admits to not working with text rendering often and invites user feedback on the matter.

  • How does the seamless pattern creation work in Midjourney v6.1?

    -In v6.1, the 'tile' parameter works better and is compatible with the new upscaler, allowing for the creation of higher resolution tiles without breaking the seamless pattern.

  • What changes were made to the model personalization system in Midjourney v6.1?

    -The model personalization system in v6.1 is more accurate and nuanced in interpreting visual aesthetic preferences based on image rankings. The personalization code changes as you rank more images, and you can see all of your personalized codes by typing '/list personalized codes' in Discord.

  • Can you use old personalization codes in Midjourney v6.1?

    -Yes, you can still use your old personalization code in both v6.0 and v6.1. You can also use any of your new personalization codes in v6.0, but you cannot combine old and new personalization codes from different systems.

Outlines

00:00

🖼️ Mid Journey v6.1 Model Overview

The video discusses the new features and improvements of Mid Journey's v6.1 model compared to v6.0. Key enhancements include better image quality with reduced pixel artifacts and improved textures, especially in eyes and skin. The coherence and accuracy of body and hand depictions have also improved. A new Q2 mode offers more texture and detail at the cost of slight coherence reduction. The video compares Mid Journey's 2X upscaler with Topaz Labs' Gigapixel, noting that Topaz provides better skin texture and eye clarity. The v6.1 model has a stronger conventional beauty bias, which was acknowledged by Mid Journey and is set to be corrected in the upcoming v7 model. The video also touches on the ability to use previous character references with the new model and the slight increase in beauty bias when using character references in v6.1. Lastly, it explores the impact of style references and SR codes in v6.1, noting a trend of simpler and less detailed images compared to v6.0.

05:00

🔍 Mid Journey v6.1 Personalization and Upscaling

The second paragraph delves into the personalization system of Mid Journey v6.1, which is designed to be more accurate in interpreting visual preferences based on user rankings. The video encourages viewers to test both old and new personalization codes to find the best fit for their needs. It also mentions the improved text rendering accuracy in v6.1 and the enhanced 'tile' parameter that maintains seamless patterns even after upscaling. The video concludes with the presenter's personal preference for the v6.0 aesthetic due to its creativity, detail, and diversity, expressing anticipation for the upcoming v7 model. The presenter also mentions upcoming releases of new SRF codes and a monthly prompt collection for Patreon members, inviting viewers to subscribe for updates and share their experiences with v6.1.

Mindmap

Keywords

💡Midjourney v6.1

Midjourney v6.1 refers to the latest version of the AI-generated image model by Midjourney, a company specializing in AI art. This version is compared to its predecessor, v6.0, in the video. The update is highlighted for its improved image quality, reduced pixel artifacts, and enhanced textures, making the generated images more realistic. It is set as the default model for all users, indicating a significant advancement in the AI's capabilities.

💡Image Quality

Image quality is a critical aspect of AI-generated art, referring to the clarity, detail, and realism of the images produced. In the context of the video, the upgrade to Midjourney v6.1 is said to bring a boost to image quality, with improvements in the clarity of eyes and skin, and better coherence and accuracy in depicting bodies and hands, showcasing the model's ability to generate more lifelike images.

💡Pixel Artifacts

Pixel artifacts are unintended visual distortions or anomalies that can occur in digital images, often resulting from compression or rendering issues. The video script mentions that Midjourney v6.1 reduces pixel artifacts, suggesting that the new model generates smoother and cleaner images compared to v6.0, which is an important step towards higher fidelity in AI art.

💡Upscaler

An upscaler is a tool or algorithm used to increase the resolution of an image or video without losing quality. In the video, the introduction of new 2X upscalers in Midjourney v6.1 is discussed, which allows for the creation of larger, more detailed images. The comparison with Topaz Labs' Gigapixel indicates that Midjourney's upscalers are competitive with other industry tools in enhancing image resolution.

💡Conventional Beauty Bias

Conventional beauty bias refers to the tendency of AI models to favor or generate images that adhere to widely accepted standards of beauty, often leading to a lack of diversity in appearance. The video points out that Midjourney v6.1 has a more pronounced conventional beauty bias, with smoother skin and fewer imperfections, which could be a limitation for users seeking more varied or unique character representations.

💡Model Personalization

Model personalization in the context of AI art generation refers to the ability of the AI to adapt and learn from user preferences to create images that align with the user's aesthetic. The video explains that Midjourney v6.1's model personalization system is more accurate and nuanced, allowing for a more personalized experience as the AI better interprets the user's visual preferences based on their image rankings.

💡SR Codes

SR codes, or style reference codes, are specific parameters used in Midjourney to guide the AI in generating images with particular artistic styles. The video discusses how SR codes can be used in both v6.0 and v6.1, but there might be differences in the level of detail and complexity of the generated images, indicating that while the style remains consistent, the rendering might vary between versions.

💡Text Rendering Accuracy

Text rendering accuracy pertains to how accurately and effectively text is incorporated into images. The video script mentions that Midjourney v6.1 should have improved text rendering accuracy, which is crucial for users who incorporate text into their AI-generated artwork, as it affects the readability and aesthetic integration of the text within the image.

💡Tile Parameters

Tile parameters are used in AI art generation to create seamless patterns that can be repeated without visible seams. The video highlights that Midjourney v6.1 fixes issues with upscaling images that break the seamless pattern, allowing for higher resolution tiles without disrupting the pattern's continuity, which is beneficial for creating patterned designs or backgrounds.

💡Character References

Character references are images or descriptions used to guide the AI in generating characters with specific features or styles. The video script notes that character references can be used with Midjourney v6.1 in the same way as with v6.0, suggesting that the AI's ability to understand and replicate character features across versions remains consistent, although the aesthetic outcome might be slightly different due to the new model's conventions.

Highlights

Midjourney's v6.1 model is now the default for all users, but can be changed in settings.

v6.1 enhances image quality with reduced pixel artifacts and improved textures.

Eyes and skin in v6.1 images are clearer and more realistic.

v6.1 improves coherence and accuracy in body and hand depictions.

The --Q2 mode in v6.1 offers more texture and details at the cost of slightly reduced image coherence.

2X upscalers in v6.1 are compared favorably with Topaz Labs' Gigapixel.

v6.1 has a more pronounced conventional beauty bias, with smoother skin and fewer imperfections.

User-generated data in v6.1 has inadvertently increased the conventional beauty bias.

v6.1 is expected to have its beauty bias corrected in the upcoming v7 model.

For unique character designs, v6.0 might be preferable over v6.1 due to less beauty bias.

Character references work similarly in both v6.0 and v6.1.

v6.1's style references show a trend towards simpler and less detailed images.

SR codes from previous versions can still be used in v6.1, with some differences.

Text rendering accuracy is reportedly improved in v6.1.

The tile parameter in v6.1 maintains seamless patterns even after upscaling.

The new model personalization system in v6.1 is more accurate and nuanced.

Personalization codes change as you rank more images in v6.1.

Old and new personalization codes can be used in both v6.0 and v6.1.

The presenter prefers v6.0 for its creativity, detail, and diversity.

v7 is anticipated to bring back some of the preferred qualities of v6.0.