The Tricks to Consistent Characters in Midjourney Version 5

The video script discusses strategies for maintaining consistent character representation across various scenes in Midjourney Version 5, a generative AI tool. The speaker debunks the myth of a built-in feature for consistency and instead offers practical tips. They emphasize the importance of a simple and human-like character design, using variations of a chosen image to guide the AI. The speaker also advises against the common misconception of 'liking' images to teach the AI preferences, which they clarify is ineffective. They then demonstrate techniques such as adjusting the weight of certain prompt elements, introducing chaos to the prompt for variety, and using negative prompts to exclude unwanted elements. The script also covers challenges when changing character poses or scenes, like buying groceries or inspecting a crime scene, and suggests using character sheets and multiple expressions for better results with animated characters. Finally, the speaker shares a trick of generating the background separately and combining it with the character prompt for more success in different settings, acknowledging the process is not perfect but can be a fun experiment.

Takeaways

• 🔍 There is no built-in feature in Midjourney that ensures consistent characters across different scenes.
• 🎭 Start by generating a human-like subject; cartoon characters are easier to maintain consistency due to their stylized nature.
• 📝 Keep your prompt simple to aid Midjourney in carrying over character design from scene to scene.
• 🖼️ Use variations on the pictures you like and upscale them for use as image prompts.
• 🚫 Disregard the myth that hitting 'favorite' on images teaches Midjourney your preferences; it has no effect.
• 🔄 Look for variations that are 94% similar but have subtle differences to provide Midjourney more to work with.
• 📌 Experiment with different prompts and scenes, adjusting the importance of elements by using weights.
• 🔩 Use the remix feature with caution as it may not produce the desired results.
• 🛍️ When facing issues with character positioning, try the slider method to adjust the ignored part of the prompt.
• 🎨 Lowering the stylized value can make Midjourney follow your prompt more closely.
• ⛓️ For animated characters, use phrases like 'character sheet' and 'multiple expressions and poses' to generate a range of images for different scenes.

Q & A

What is the main challenge when trying to get consistent characters across different scenes in mid-journey?

The main challenge is that there is no built-in feature in mid-journey that allows for consistent character generation across different scenes. It requires using various tricks and techniques to achieve a semblance of consistency.

Why is it recommended to prompt a human-like subject when generating characters?

Photorealistic humans tend to look the same, whereas a cartoon or a stylized character can be drawn in many different styles, making it easier to maintain consistency across scenes.

What is the first trick mentioned for generating consistent characters?

The first trick is to keep the design simple. A simple prompt is more likely to be carried over from scene to scene by the mid-journey system.

What is the significance of using variations on the pictures you like?

Using variations provides more than one picture to use as an image prompt, which can help in getting a more consistent character across different scenes.

Why is hitting 'favorite' on images considered a bad advice in mid-journey?

Hitting 'favorite' is a placebo effect and a myth. It does not teach mid-journey what the user wants, and it has no impact on the image generation process.

How can the remix feature in mid-journey be used to create variations?

The remix feature allows you to take a picture you like and attempt to remix it into something else, such as changing the pose or adding the character into a different scene. However, the results may not always be as expected.

What is the purpose of adding weights to the prompt in mid-journey?

Adding weights to the prompt helps to prioritize certain aspects of the image generation. For example, making 'sitting on a park bench' 1.5 times more important than the reference images can influence the outcome to better match the desired scene.

What does adding chaos to the prompt in mid-journey do?

Adding chaos to the prompt increases the variety of the grid given to the user. It can help in finding a good generation among multiple attempts, but it also risks straying from the prompt significantly.

Why might using the slider method not always work when trying to change the scene of a character?

The slider method may not work if the reference images are too specific or limited in pose, as the system tends to generate images that are similar to the reference images, making it difficult to introduce new scenes or poses.

What is the 'character sheet' trick for generating animated or cartoon characters?

The 'character sheet' trick involves using phrases like 'multiple expressions and poses' in the prompt to generate a variety of character images that can then be separated and placed into different scenes using image editing software like Photoshop.

What is the final trick suggested for generating a consistent character in different scenes?

The final trick is to generate the background separately and then include that in the image prompts. This can help in creating a more natural and integrated scene for the character.

Outlines

00:00

🎨 Maintaining Character Consistency in Mid-Journey

The paragraph discusses the challenge of maintaining consistent character designs across different scenes in the AI art generation tool, mid-journey. It dispels the myth of a built-in feature for consistency and instead offers practical tips. The speaker emphasizes the importance of simplicity in design and suggests using variations of a preferred image to guide the AI. They also caution against the misconception that favoriting images influences the AI's future outputs, labeling it as an urban legend. The paragraph concludes with a demonstration of how to use multiple images and prompts to achieve a more consistent character representation.

05:01

🔍 Refining the AI's Output with Weights and Chaos

This section delves into techniques for refining the AI's generated images to better match the user's desired outcome. It explains the use of weights, which allow certain aspects of the prompt to be emphasized over others. The paragraph also introduces the concept of adding 'chaos' to the prompt to increase the variety of the generated images, potentially leading to more satisfactory results. However, it warns against setting the chaos level too high, as this can lead to outputs that deviate significantly from the prompt. The speaker shares their trials with different prompts and settings, illustrating the hit-or-miss nature of the process.

10:01

🚧 Overcoming Limitations in Scene Generation

The final paragraph addresses the limitations when trying to place a consistent character into various scenes. It describes several strategies, such as using a 'character sheet' prompt for more animated or cartoon-like characters, and the slider method to adjust the importance of certain prompt elements. The speaker also discusses the use of negative prompts to exclude unwanted elements, like the camera lens appearing in the generated images. The paragraph concludes with a demonstration of generating a background separately and combining it with a character to create a more cohesive scene, although it acknowledges the inherent challenges and imprecision in this approach.

Keywords

💡Mid-journey

Mid-journey refers to a version of a generative art tool that creates images based on user prompts. In the video, the speaker discusses techniques to maintain character consistency across different scenes using this tool, highlighting that while there is no perfect solution, several tricks can be employed to achieve a semblance of consistency.

💡Consistent Characters

Consistent characters refer to the ability to depict a character with the same visual and stylistic attributes across different scenes or images. The video emphasizes the challenge of achieving this consistency in generative art tools like Mid-journey, and provides strategies to get as close as possible to this ideal.

💡Human-like Subject

A human-like subject in the context of the video refers to a character or figure that resembles a human being, but is stylized in a way that allows for more variation and uniqueness compared to photorealistic humans. The speaker suggests using cartoon or stylized humans to have more control over the character's appearance in different scenes.

💡Prompt

In the context of the video, a prompt is a text input provided by the user to guide the Mid-journey tool in generating a specific image. The speaker discusses the importance of crafting simple and effective prompts to better control the outcome and maintain character consistency.

💡Variations

Variations in this context refer to the different images generated by Mid-journey based on the same prompt, which can be used to find subtle differences that can help in achieving consistency across scenes. The speaker advises using these variations as a reference for future prompts.

💡Weights

Weights in the video script refer to a method of emphasizing certain elements of the prompt over others when generating images. By adding a weight (indicated by double colons followed by a number), the user can increase the importance of a specific part of the prompt, which can help in getting a more accurate representation of the desired character or scene.

💡Chaos

Chaos in the video is a parameter that introduces variability into the generated images. By adjusting the chaos value, the user can control the level of randomness in the output, which can be useful for finding a good image among several attempts, but it should be used carefully to avoid straying too far from the desired result.

💡Character Sheet

A character sheet is a concept used in the video to refer to a collection of images that show different expressions and poses of a character. This technique is particularly useful for animated or cartoon characters, where a variety of expressions and poses can be combined and placed into different scenes.

💡Background

In the video, the term 'background' refers to the setting or environment in which the character is placed. The speaker discusses generating backgrounds separately and then combining them with the character images to create a more natural and candid look in different scenes.

💡Slider Method

The slider method is a technique mentioned in the video for adjusting the emphasis on certain parts of the prompt. By moving a part of the prompt to the end and adding a weight, the user can influence the generation to focus more on that particular aspect, which can help in achieving the desired pose or scene.

Highlights

There is no built-in feature in Midjourney that allows for consistent character generation across different scenes.

Prompting a human-like subject can lead to more consistent results as cartoon characters can be drawn in various styles.

Keeping the design and prompt simple helps maintain consistency from scene to scene.

Using the word 'detective' can influence the fashion style of the character.

Adding specific details like hair color and style to the prompt can be beneficial.

Variations on preferred pictures can be used as an image prompt to achieve more consistency.

Hitting 'favorite' on images does not affect the AI's learning or future image generation.

Looking for variations that are 94% the same but have subtle differences can improve results.

The remix feature in Midjourney does not work as expected for changing poses or scenes.

Using multiple reference images and copying their addresses into the prompt can enhance consistency.

Experimenting with different words, scenes, and mixing up the prompt can lead to better results.

Placing important keywords at the end of the prompt or using weights can emphasize certain aspects.

Adding chaos to the prompt can increase the variety of the grid, potentially yielding better images.

The slider method can help when a part of the prompt is being ignored by moving it to the end with added weight.

Lowering the stylized value makes Midjourney follow the prompt more closely.

Using negative prompts, such as 'no camera lens', can help remove unwanted elements from the image.

The character sheet method with multiple expressions and poses works better for animated characters.

Generating the background separately and including it in the image prompt can improve scene consistency.

Using the blend feature with two pictures can sometimes yield better results.

Being specific in the prompt, such as requesting a side profile, can lead to more accurate character depiction.

Consistent character generation in different scenes is challenging and requires a lot of trial and error.