Midjourney Slider Method: How to fine-tune images by Multiprompting
Fine-tune your image like a Pro
The Slider Method is a technique to fine-tune an image with multiprompting. The first promptlet’s words are used to create other promptlets, which are then weighted to adjust the final image.
If you’re new to multiprompting, you can learn more about it in this story.
The Slider Method can help you to do the following:
(1) Adjust the influence of the element(s) in your prompt.
- Numerical adjustment
- Descriptive adjustment
- Implied adjustment
(2) Force an element to be expressed in your image when the Midjourney bot ignores it. This is useful for troubleshooting prompts.
(3) Remove a long list of elements that you do not want to see in the image.
Point (1) is covered in great detail in this story. Other topics will be addressed in another story.
The promptlets must be given a weight to use the Slider Method.
Here are the key points about multi-prompt weight assignment:
- Each prompt is controlled by adding a weight value.
- First, try not to add weight to the promptlet and see if it works for you. Often, it works just fine without adding any weight. There is no need to rush to add weight unnecessarily.
- Lower weight has less influence, while higher weight has more. The influence of the promplet contributes to the final blended output.
- The default value is 1. If you don’t explicitly assign a weight, the bot will automatically set it to 1.
- A negative value indicates that the element(s) in that promptlet must be removed from the final blend.
- The weight value has 2 decimal points.
- You can calculate the total prompt weight of a multi-by by adding the weights of the promplets. The total weight cannot be zero or negative.
- If the total weight value is negative, the Midjourney bot will pop up a message to warn and stop you. (Quite fun to trigger the bot; you can try it yourself later).
- If the total weight value is near zero, the bot will “go crazy’ and generate strange images.
- Ideally, and to be safe, adjust the total weight to at least greater than 0.3.
- The official recommended incremental positive values that you can assign to a promptlet are 0.25, 0.3, 0.5, 0.6, and 0.7.
- The recommended incremental negative values are the reversal of positive values: -0.7, -0.6, -0.5, -0.3, and -0.25. (Typically, you will only use the -0.5 value because it’s equivalent to the
--noparameter) - You can also use other values other than the officially recommended incremental values, like 0.2 or 1.6. It’ll work just fine.
- The influence of the assigned weights of a promplet is relative to the weights of other promptlets. Based on the Rules of Weighted Average, these promptlets have the same influence.
/imagine prompt: promptlet A::5 promptlet B::10is the same as/imagine prompt: promptlet A::1 promptlet B::2 - There’s a word limit for a multi-prompt. In V5.2, each multi-prompt can hold up to about 60 tokens.
- The Midjourney bot will break up the words in your prompt into pieces known as “tokens” when it reads them. Usually, the number of tokens equals or exceeds the number of words. For example, a multi-prompt with 34 words could have 34 or more tokens.
- You don’t have to check the number of tokens when using multi-prompt. It’s FYI only.
- Using the Tokenizer, you can check how many tokens are in your multi-prompt. It’s not an exact science experiment; it’s just an estimate.
- Use this Tokenizer as recommended by Midjourney: https://novelai.net/tokenizer
- Choose CLIP Tokenizer from the dropdown menu.
- When you use the tool, remove any parameters, like
--ar 3:2.
If you read through the 15 points above, you’re pretty awesome. I hope it’s not too frightening.
That’s the reason why in the previous story, I said that the multi-prompt topic is challenging and complicated to write about.
Adjust the influence of the element(s) in your prompt
(a) Numerical adjustment
You can change an element’s influence by creating a promptlet with the exact word(s) from the first promptlet. Then, to adjust its influence, assign a weight.
/imagine prompt: a happy chicken in the field
/imagine prompt: a happy chicken in the field:: happy chicken::
/imagine prompt: a happy chicken in the field:: happy chicken::2
The first prompt is the prompt without a multi-prompt.
The second prompt contains a promplet that repeats the word “happy chicken”. If no weight is explicitly assigned, the weight value for happy chicken:: is set to 1.
That is, the second prompt is actually read as follows:
/imagine prompt: a happy chicken in the field::1 happy chicken::
You can see that the happy effect is more obvious than the first prompt.
The “happy chicken” promplet is set to 2 in the last promptlet. It’s no surprise that the chickens are ecstatic.
(b) Descriptive adjustment
You can also use non-numerical values to adjust the slider. A “slider” is the word(s) you chose from the first promptlet to create another promptlet where you can adjust its value.
You can make chickens with shades of green by assigning different color descriptions.
And, of course, you can combine numerical and descriptive adjustments together.
/imagine prompt: a happy chicken in the field:: green chicken::
/imagine prompt: a happy chicken in the field:: dark-green chicken::
/imagine prompt: a happy chicken in the field:: dark-green chicken::2
(C) Implied adjustment
The official Slider Method recommendation is to repeat the exact word(s) from the first promplet. Otherwise, it won’t work.
But, I have found another way to extend the method.
Creating a slider without repeating the exact word is possible.
If the composition is simple, such as a person’s portrait, we can directly add detail like an adjective without repeating the word from the first promptlet. The Midjourney bot will apply the adjective to the image accordingly.
Prompt 1
/imagine prompt: a close-up photo portrait of a young man
Prompt 2
/imagine prompt: a close-up photo portrait of a young man:: wrinkle::1
Prompt 3
/imagine prompt: a close-up photo portrait of a young man:: wrinkle::0.7
Prompt 4
/imagine prompt: a close-up photo portrait of a young man:: a mole on the forehead::
Prompt 1 is the base prompt, nothing fancy.
Prompts 2 and 3 add the word “wrinkle” which is not found in the first promptlet. You can see the wrinkle effect became significantly lesser as the weight was reduced from 1 to 0.7.
Prompt 4 shows that it’s also possible to blend in other detail, in this case, a mole, to the forehead of the subject. Can you find the mole?
Prompt 1
/imagine prompt: a studio photo of a Chihuahua:: white
Prompt 2
/imagine prompt: a studio photo of a Chihuahua:: white:: hair::
Prompt 3
/imagine prompt: a studio photo of a Chihuahua:: white:: hair::2
Prompt 4
/imagine prompt: a studio photo of a Chihuahua:: white:: Chihuahua hair::2
The implied adjustment technique can occasionally generate unexpected photos. You can correct this by repeating the word from the first promptlet.
Prompts 1 and 2 work as expected. Prompt 3 creates an odd photo in which the dog is missing, and there’s a long-haired lady. This problem can be fixed by reintroducing a word (Chihuahua) from the first promptlet.
I hope you find the Slider Method useful. I use it quite a lot in my Midjourney projects.
Related stories
Practical use of the Slider Method in Vary (Region) inpainting.
Level up your prompting skill with multiprompting.
An overview of Midjourney’s layout and composition techniques.
Crafting a good prompt is both an art and a skill.
Conclusion
- The Slider Method is a multiprompting method that uses promptlets to adjust an image. The words from the first promptlet are used to create another promptlet(s) and given weight.
- When giving the promptlet a weight, the factors affecting the final image should be considered.
- You can change a promptlet’s influence by giving it a number (numerical adjustment), making it more information (descriptive adjustment), or adding a new adjective that is not found in the first promptlet (implied adjustment). You can also make all of these fine-tuning simultaneously.
I hope you like this story!
Please give me a clap or leave a positive comment to let Medium’s algorithm know that this story is helpful and has contributed to its platform’s success.
Your support inspires me to create similar or higher-quality content in the future!
Never miss a story. Follow me on Medium for updates, stories, and tips about Midjourney.
Join Medium today and read all my stories:
