【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. The feature, called Mijurney character reference (CREF), works by referencing images through URLs and adjusting the weight of the reference with the CW parameter. It's best used with MIGI6 or MIGI6 models and is particularly effective for anime-style characters. The video demonstrates how to use CREF and CW to alter a character's clothing while retaining facial features, showcasing the feature's potential and limitations.
Takeaways
- 🌟 A new feature has been released for AI consistency in character generation, allowing for the stable creation of the same character.
- 🚀 The feature is particularly beneficial for users who previously found control over character generation to be lacking.
- 📌 The Mijurney character reference (CREF) and CW (character weight) parameters have been introduced for fine-tuning character features.
- 🔗 CREF requires a URL of an image to be used as a reference for character generation.
- 📊 The CW parameter adjusts the weight of the reference image, with a range from 100 to 0, defaulting to 100 for detailed reference.
- 🎨 Lower CW values reduce the detail of the reference, allowing for changes in aspects like clothing and hair while keeping the facial features consistent.
- 🌟 The feature works best with characters generated within the Mijurney platform, but it also performs well with other sources.
- 🛠️ The precision of the feature is limited and cannot perfectly replicate intricate details like moles or logos.
- 🤖 The CRIB function supports both Niji journey and Mi journey models but requires the use of the MIGI6 model.
- 🎭 Users can combine the new feature with the previously released Swift function for enhanced character generation.
- 🌐实操 demonstration is provided to show the effectiveness of the CREF and CW parameters in character customization.
Q & A
What new feature was announced at 6 AM that everyone has been eagerly waiting for?
-The new feature announced is AI character consistency, which allows for the stable generation of the same character.
How does the Mijurney character reference (CREF) feature work?
-The Mijurney character reference (CREF) feature works by allowing users to reference a character's features using a URL of an image after adding a specific parameter (杠杠CREF) in the input text.
What is the purpose of the 杠杠CW parameter?
-The 杠杠CW parameter adjusts the weight of the reference URL, determining how much detail from the reference image is incorporated into the generated image. The value ranges from 100 to 0, with higher values leading to more detailed references.
What is the recommended model version to use with the CREF feature?
-The recommended model version to use with the CREF feature is MIGI6, as it currently only supports this version and not MIGI5.2 or earlier versions.
What are the benefits of using the CREF feature with Niji journey or Mi journey models?
-Using the CREF feature with Niji journey or Mi journey models allows for the generation of characters with more consistent facial features and improved overall character recognition from reference images.
How does the precision of the CREF feature currently limit the details it can replicate?
-The precision of the CREF feature is currently limited and cannot fully replicate intricate details such as dimples, freckles, or logos on t-shirts.
What happens when the 杠杠CW parameter is set to 0?
-When the 杠杠CW parameter is set to 0, the generated image will primarily focus on replicating the facial features and will not significantly alter other aspects like clothing or hair.
How does the use of the ITA tag affect the generation process?
-The ITA tag is used to specify the subject in the generation process, ensuring that the AI recognizes and incorporates the specified elements into the generated image.
What was the tribute made at the end of the script?
-The tribute was made to the creator of Dragon Ball, Akira Toriyama, expressing respect and怀念 for the creator of a cherished childhood memory.
How did the demonstration of the CREF feature utilize the character Uchiha Itachi?
-The demonstration used Uchiha Itachi as a reference character, showing how the CREF feature could change his clothing while maintaining his facial features and other distinctive attributes.
What was the unexpected result when the ITA tag was omitted and only the CREF feature was used?
-When the ITA tag was omitted and only the CREF feature was used, the AI generated an image of a woman, possibly due to the influence of the 'pink' element in the input, showing that the AI can still generate consistent facial features without a specific character name.
Outlines
🚀 Introduction to AI Character Consistency Feature
The paragraph introduces a new feature for AI character consistency, which allows for the stable generation of the same character. It discusses the excitement around this release, particularly for the AI known as me Joni, which has been criticized for lack of control in the past. The video aims to explain the Mijurney character reference (CREF) feature and its usage, comparing it to the previously released style reference feature. The CREF function involves adding a URL after a specific parameter, which is demonstrated practically in the video. It also mentions the new parameter 'CW' for adjusting the weight of the reference URL, with a range from 100 to 0, affecting the level of detail in the character's features. The feature is particularly effective when using images generated by the Mijurney model itself. The limitations of the feature, such as the inability to perfectly replicate details like dimples or logos, are acknowledged, but the potential for future improvements is highlighted. The CREF function is noted to support both Niji journey and me journey models but is only compatible with the MIGI6 model.
🎨 Demonstrating the CREF and CW Parameters
This paragraph continues the discussion on the CREF feature by providing a practical demonstration of its use. It showcases the process of generating images using the CREF and CW parameters, starting with selecting a character, in this case, Uchha Itoch from the anime. The video demonstrates how different CW values (0, 50, and 100) affect the generation, with varying levels of detail in the character's clothing and facial features. The paragraph also explores the effects of using CREF without specifying the character's name, showing how the AI can still generate a consistent facial appearance based on the reference image. The results of these tests are presented, highlighting the versatility and precision of the CREF feature in creating characters with specific attributes and clothing. The paragraph concludes with a tribute to the creator of Dragon Ball, Akira Toriyama, and a reminder to use the MIGI6 model for the best results.
Mindmap
Keywords
💡AI角色一致性
💡Mijurney character reference (CREF)
💡杠杠CW (CW parameter)
💡Miji
💡Niji journey
💡Swift function
💡MIGI6
💡权重 (Weight)
💡写实人物 (Realistic characters)
💡宇智波鼬 (Uchiha Itachi)
💡鸟山明 (Akira Toriyama)
Highlights
Introduction to the highly anticipated feature of AI character consistency, enabling stable generation of the same character.
Overview of Mijurney character reference CRIF, the latest feature for character consistency.
Explanation of how to use the CRIF feature by adding a URL after the parameter in your command.
Introduction of the new parameter CW (Character Weight), which adjusts the weight of the character reference URL.
Details on how the CW parameter affects character generation, with a scale from 0 to 100 influencing the detail level of the generated image.
Tips on using the CW parameter for altering specific character features while maintaining facial consistency.
Insight into the optimal usage of the feature with characters generated within the Miji ecosystem for best results.
Discussion on the limitation of precision in replicating minute details like dimples, freckles, or T-shirt logos.
Confirmation that the CRIB feature is supported by both Niji Journey and Me Journey models.
Information on the possibility of using multiple URLs as references for character generation.
Real-world application demonstration using Nigi Journey to generate a character with distinct facial and clothing features.
Testing the effectiveness of different CW values in altering a character's appearance while maintaining facial details.
A comparative analysis of images generated with different CW values to showcase the versatility of the feature.
Demonstration of the feature's ability to adapt the gender of a character based on the input parameters and references used.
Tribute to Akira Toriyama, expressing gratitude for his contribution to the presenter's childhood through Dragon Ball.
Highlight of the feature's potential error handling when incorrect references are used.