【AI角色一致性】“cref” 参数 Midjourney新功能实操详解
TLDRThe video introduces a new AI feature for character consistency, allowing stable generation of the same character. It explains the use of the Mijurney character reference (CREF) and the CW parameter for adjusting the weight of the reference image. The feature is best used with the MIGI6 model and can be combined with the Swift function for enhanced results. The video demonstrates the feature using an example of changing Uchiha Itachi's outfit while retaining his facial features, showcasing the effectiveness of the CREF and CW parameters in generating consistent character images.
Takeaways
- 🚀 New Feature Launch: A new feature has been released that ensures consistency in AI character generation, allowing for the stable creation of the same character.
- 🔄 Character Consistency: The feature addresses previous concerns about the lack of control over character consistency, significantly improving user experience.
- 📸 Image Reference (CREF): Users can now reference character features using URLs, which is similar to the previously introduced style reference feature but focuses on character traits.
- 🔗 URL Usage: The CREF parameter requires a URL to be placed after it, which should be a link to an image that the user wishes to use as a reference for character generation.
- 📊 Weight Adjustment (CW): The new parameter, CW, allows users to adjust the weight of the reference URL, with a range from 100 to 0, affecting the level of detail in the generated image.
- 🎨 Detail Preservation: A higher CW value (closer to 100) preserves more details such as facial features, hair, and clothing, while a lower value (closer to 0) focuses primarily on facial details.
- 🌟 Optimal Model: The feature works best with the Mijurney character reference, especially when the reference is from images generated within the Mijurney model itself.
- 📷 Image Precision Limitations: While the feature is powerful, it currently has limitations in precision, such as not being able to replicate small details like dimples or freckles perfectly.
- 🤖 Compatible Models: The CRIB function supports only MIGI6 or MIGI6 models, and not older versions like MIGI5.2.
- 🌐 Multiple URL Support: Users can use one or multiple URLs as character references, enhancing the precision and customization of the generated character images.
- 🎭 Transformative Potential: The feature enables the transformation of realistic characters into anime styles with improved facial effects, thanks to the new level of control over character features.
- 🕊️ Tribute to Creators: The script includes a tribute to the creator of Dragon Ball, Akira Toriyama, and the impact of his work on the speaker's childhood and the broader community of fans.
Q & A
What new feature was announced at 6 AM that everyone has been eagerly awaiting?
-The new feature announced is AI character consistency, which allows for the stable generation of the same character.
What is the significance of the character reference feature (CREF) introduced in the update?
-The character reference feature (CREF) is significant because it enables users to reference specific character traits, allowing for more controlled and detailed generation of characters in images.
How is the CREF parameter used?
-The CREF parameter is used by adding a URL of an image after the 'bar' (-) symbol followed by 'CREF'. This URL is used as a reference for the character's features.
What is the purpose of the 'bar' (-) CW parameter?
-The 'bar' (-) CW parameter adjusts the weight of the reference URL, similar to the image weight parameter in previous functions. It determines how much of the reference image's details are incorporated into the generated image, with a range from 100 (most details) to 0 (only facial features).
Which model versions support the CREF feature?
-The CREF feature supports MIGI6 and MIGI6 models, but not MIGI5.2 or earlier versions.
What is the best practice when using the CREF feature for optimal results?
-For optimal results, it is recommended to use the CREF feature with the Miji or Niji journey models and to provide a clear and detailed reference image that closely matches the desired character traits.
What are the limitations of the CREF feature in terms of detail reproduction?
-The CREF feature currently has limited precision and cannot perfectly reproduce intricate details such as dimples, freckles, or logos on t-shirts.
How does the CREF feature work in conjunction with other parameters?
-The CREF feature can be used alongside other parameters, such as the 'ITA' (item tag attribute) parameter, to further refine the character generation process.
What was the example character used in the script to demonstrate the CREF feature?
-The example character used to demonstrate the CREF feature was Uchiha Itachi from the Naruto series.
How did the different CW values affect the generation of Uchiha Itachi with a white shirt?
-CW values of 0, 50, and 100 resulted in different levels of detail from the reference image. At CW 0, the face was retained with minimal changes to clothing. At CW 50, both the face and some clothing details were retained. At CW 100, the reference image's details were heavily incorporated, resulting in minimal changes to the clothing and a more detailed facial reproduction.
What was the unexpected outcome when the 'Itachi' tag was removed from the CREF command?
-When the 'Itachi' tag was removed, the AI generated an image of a woman, possibly due to the influence of the 'pink' element in the command, indicating that the absence of a specific character tag can lead to different interpretations by the AI.
Outlines
🚀 Introduction to AI Character Consistency Feature
The paragraph introduces a newly released feature for AI character consistency, which allows for the stable generation of the same character. The speaker, Joni, expresses excitement about this feature that has been highly anticipated and shares a brief test result. The video's main focus is to explain the 'Mijourney character reference (CREF)' feature and its usage, including how to use it with a URL for character reference and the 'CW' parameter to adjust the weight of the reference. The speaker also mentions that this feature works best with characters generated in Miji and is currently only supported by the MIGI6 model.
🎨 Demonstrating the CREF and CW Parameters
This paragraph details the practical application of the CREF and CW parameters. The speaker demonstrates how to use these parameters to change a character's clothing while retaining facial features, using Uchiha Itachi as an example. The paragraph explains the different effects of varying the CW parameter values (0, 50, and 100) and how they impact the character's appearance. The speaker also explores the results of using CREF without specifying the character's name, showing how AI can still generate a consistent facial appearance based on the reference image. The paragraph concludes with a tribute to the creator of Dragon Ball, Akira Toriyama, and a note on the importance of using the correct parameters for the best results.
Mindmap
Keywords
💡AI角色一致性
💡Mijurney character reference (CRIF)
💡杠杠CW (CW parameter)
💡Miji
💡Niji journey
💡Swift function
💡MIGI6
💡Uchha Itoch
💡ITA
💡Segret
💡鸟山明
Highlights
AI角色一致性功能发布，允许稳定生成同一个角色。
Mijurney character reference (CRIF)功能介绍，用于增强角色特征的一致性。
CRIF功能类似于之前的style reference功能，但专注于人物特征。
使用CRIF时，通过在关键词后添加URL来引用角色特征。
新参数杠杠CW用于调整参考网址的权重，影响角色特征的还原程度。
CW参数的预值范围为100到0，数值越高，参考的细节越多。
CRIF功能在me journey和Miji journey模型上效果最佳。
CRIF功能目前只支持MIGI6模型，不支持MIGI5.2等旧版本。
CRIF功能可以与Swift功能结合使用，提高角色生成的精确度。
CRIF功能通过URL引用角色特征，支持一个或多个URL。
使用CRIF功能时，面部特征如护额、写轮眼等可以得到很好的保留。
CRIF功能允许在保持面部不变的情况下，改变角色的穿着和发型。
CRIF功能在处理写实人物转动漫风格时，面部效果有显著提升。
CRIF功能的强大之处在于即使不加主体名称，也能通过参考图生成一致性角色。
CRIF功能的工作原理与正常做图效果相似，但更聚焦于角色特征识别。
CRIF功能目前精度有限，无法完全还原如酒窝、雀斑等细节。
向鸟山明老师致敬，表达了对经典漫画的热爱和对童年英雄的怀念。