HUGE Midjourney Update! Model Personalization, Is It Good? | Midjourney v6 Tips

Woollyfern
14 Jun 202404:22

TLDRMidjourney introduces a new model personalization feature, allowing users to create a personalized algorithm based on their aesthetic preferences. By ranking at least 200 image pairs, users can teach Midjourney their style, which is then applied through a personalized parameter in prompts. This feature offers a diverse set of creative possibilities, enabling users to escape the default bias and explore a more tailored visual space. The video provides tips on using this feature and encourages viewers to share their personalized codes.

Takeaways

  • 🔄 Midjourney has introduced a new feature called 'Model Personalization' that allows users to have their own personalized algorithm.
  • 🎨 Each Midjourney model has a default bias and style aesthetic that affects the color palettes and subject matter in generated images.
  • 🛠 To counteract the default style, users can now use their personal bias and style aesthetic with the new feature by adding the parameter '--p' to their prompt.
  • 📈 The personalized model is trained by ranking at least 200 image pairs on the website, which teaches Midjourney the user's style preferences.
  • 📸 Users can rank image pairs by choosing the one they prefer, regardless of prompt accuracy, to indicate their personal style direction.
  • 📈 Once 200 image pairs have been ranked, the personalized model is ready to use, and can be invoked with the '-p' parameter in prompts.
  • 🔑 Each Midjourney user has a unique model personalization code that can be shared with others to try out their personalized style.
  • ⚙️ The strength of the personalized model can be adjusted using the 'stylize' value, with a default of 100 and a range from 0 to 1000.
  • 🔍 Model personalization can reveal a lot about an individual's style preferences and can be tested with a variety of prompts for comparison.
  • 🌐 The personalized model can be used in conjunction with other Midjourney features like character and style references for further exploration.
  • 🔄 Users have the option to set their personalized model as the default in their settings for ongoing use.

Q & A

  • What is the new feature introduced in Midjourney called?

    -The new feature introduced in Midjourney is called 'Model Personalization'.

  • How does the model personalization feature in Midjourney work?

    -The model personalization feature allows users to access a personalized Midjourney algorithm that aligns with their own aesthetic preferences. Users can try their personalized model by adding the parameter '--p' after their prompt.

  • What is the default bias and style aesthetic in each Midjourney model?

    -Each Midjourney model has a default bias and style aesthetic that impacts everything from color palettes to the subject matter in the images generated.

  • How can users fight against the default Midjourney bias and style?

    -Users can fight against the default bias and style by using various approaches, and with the new model personalization feature, they can teach Midjourney their personal bias and style aesthetic.

  • What is required to use the model personalization feature?

    -To use the model personalization feature, users need to rank at least 200 image pairs on the Midjourney website, which helps the algorithm learn the user's style preferences.

  • How does the ranking process on Midjourney's website work?

    -The ranking process involves going to the ranking page and for each pair of images, clicking the one that the user likes better, regardless of whether the image is true to the prompt.

  • What happens when a user has ranked 200 or more image pairs?

    -Once a user has ranked 200 or more image pairs, they can start using their personalized model by including '-d-p' at the end of their prompt.

  • Are the model personalization codes unique to each user?

    -Yes, each Midjourney user has their own unique model personalization code, and these codes are sharable.

  • How can the strength of the personalized model be adjusted?

    -The strength of the personalized model can be adjusted by changing the 'stylize' value, which has a default value of 100 and can range from zero to 1,000, with 1,000 being the maximum strength.

  • What kind of insights can Midjourney's model personalization reveal about a user's style preferences?

    -Midjourney's model personalization can reveal a lot about a user's style preferences, for better or for worse, by generating images that align with the user's ranked image pairs.

  • Can the personalized model be used with other Midjourney features like character and style references?

    -Yes, the personalized model can be used with features like character references and style references, opening up a world of exploration beyond the default Midjourney bias.

Outlines

00:00

🎨 Mid Journey's Model Personalization Feature

Mid Journey introduces a new model personalization feature that allows users to create a personalized algorithm based on their aesthetic preferences. The default bias and style of each model affect various aspects of image generation, including color palettes and subject matter. Users can counteract the default style by using a personalized model, which is activated with the '--p' parameter in prompts. To train the model to recognize personal style, users must rank at least 200 image pairs on the website, indicating preferences by selecting the more appealing image. This process teaches Mid Journey the user's style direction, and the more selections made, the better the model learns the user's preferences. Once the ranking threshold is met, users can employ their personalized model in image generation, and the strength of personalization can be adjusted with the 'stylize' value. The feature not only reveals insights into an individual's style preferences but also offers a customizable experience that diverges from the default settings.

Mindmap

Keywords

Midjourney

Midjourney refers to a specific AI model or software that generates images based on textual prompts. In the context of the video, it discusses an update to the Midjourney model that allows for personalization, which is a significant development in AI-generated art and design. The script mentions how the default bias and style aesthetic of each Midjourney model can be overridden with a personalized model, which is a key feature of the update.

Model Personalization

Model Personalization is the ability to tailor an AI model to an individual's preferences. In the video, it is presented as a new feature of the Midjourney model, allowing users to create a version of the AI that aligns with their own aesthetic and style preferences. This is achieved by ranking image pairs on the website, which teaches the AI the user's style, as demonstrated in the script with the personalized parameter --p.

Default Bias

Default Bias in the context of AI image generation refers to the inherent tendencies or preferences built into the model's algorithm. For example, the script mentions that prompting for a photo of a person might default to generating an image of a woman. This bias can be influenced or overridden through model personalization.

Style Aesthetic

Style Aesthetic pertains to the visual characteristics and preferences that define an individual's or a model's artistic expression. The video discusses how Midjourney's model personalization feature allows users to imprint their own style aesthetic onto the AI, affecting aspects like color palettes and subject matter in the generated images.

Prompt

A prompt in AI image generation is a textual description or instruction given to the AI to guide the creation of an image. The script explains how different prompts can elicit different responses from the AI, and how the new model personalization feature changes the way users prompt Midjourney to generate images that match their personal style.

Ranking Image Pairs

Ranking Image Pairs is the process described in the script where users select which of two images they prefer. This selection process is how Midjourney learns the user's style preferences. The more image pairs a user ranks, the better the AI understands and adapts to the user's aesthetic, as illustrated by the requirement to rank at least 200 image pairs.

Personalized Model

A Personalized Model, as discussed in the video, is a version of the Midjourney AI that has been trained to reflect an individual user's style and preferences. It is created through the ranking of image pairs and can be used by adding a specific parameter to the user's prompts, as shown in the script with the example of adding --p to the prompt.

Stylize Value

The Stylize Value is a parameter that can be adjusted to increase or decrease the strength of the personalized model's influence on the generated images. The script mentions that the default value is 100, with a range from zero to 1,000, where 1,000 represents the maximum strength of the personalization.

Seed Number

A Seed Number in AI image generation is a specific value that, when used, ensures the reproducibility of the image generation process. The script suggests using a seed number to run the same prompt with and without model personalization to compare the results directly, providing a method to analyze the impact of personalization on the AI's output.

Character References

Character References are additional inputs provided to the AI to guide the generation of images that include specific characters or traits. The script mentions that the personalized model can be used in conjunction with character references, allowing for more customized and targeted image generation.

Style References

Style References are inputs similar to character references but focus on the artistic style or aesthetic elements to be included in the image. The video script explains that, like character references, style references can be used with the personalized model to further tailor the AI's output to the user's preferences.

Highlights

Midjourney introduces a new model personalization feature allowing users to have their own personalized algorithm.

Each Midjourney model has a default bias and style aesthetic that affects image generation.

The new feature enables personalized image generation by adding the parameter '--p' to prompts.

Personalized models can be achieved by ranking at least 200 image pairs on the Midjourney website.

Ranking image pairs helps Midjourney learn the user's style preferences.

The more image pairs ranked, the better Midjourney understands the user's aesthetic direction.

Once 200 image pairs are ranked, users can start using their personalized model with the 'd-p' parameter.

Personalization codes are unique to each user and can be shared with others.

The strength of the personalized model can be adjusted with the 'stylize' value, ranging from 0 to 1000.

Model personalization can reveal insights into a user's style preferences.

Testing personalized models with diverse prompts can demonstrate their adaptability.

Comparing results with and without model personalization can highlight differences in image generation.

Personalized models can be used alongside features like character and style references.

The feature offers a way to escape Midjourney's default bias and create more tailored image outputs.

Users can enable model personalization by default in their settings for consistent use.

The video encourages viewers to like, subscribe, and comment for more Midjourney content.

The speaker shares their personal bias towards teal, black, moody, misty, and sometimes strange images.

The video provides a direct comparison between the default model and the speaker's personalized model using the same seed approach.