MidJourney: Multiprompting
Introduction[edit | edit source]
In V3 it was already possible to multiprompt and V4 reintroduces the feature. Instead of entering a single prompt you will be giving the Bot multiple prompts in one command.
Single prompt: /imagine A cat and a block of cheese --v 4
Multiprompt:
/imagine A cat:: a block of cheese
or (longer notation; same meaning)
/imagine A cat::1 a block of cheese::1
You decide where the next prompt start with the double :: followed by a number. Leaving the number blank will be interpreted as 1.
1. Weighting[edit | edit source]
The numbers behind the :: is the weight of that particular prompt. So back to our example:
/imagine A cat::1 a block of cheese::1
Both prompts are equally important, but you can play with this. Weights can even be negative if you would like something to NOT be there.
/imagine A cat::1 a block of cheese::2 --> makes the cheese twice as important as the cat
/imagine A cat::2 a block of cheese::1 --> makes the cat twice as important as the cheese
There are a few rules:
- Its about relative weight compared to each other; the total amount doesn't really matter
- Total of prompt weights must be positive
- Total of prompt weight can be 0; but the results will be chaotic and unpredictable (and possibly fun, but mostly garbage)
2. What happens when you use multiprompts[edit | edit source]
As said in the introduction when you use multiprompting multiple prompts will be sent to the model. When resolved these will be combined into a single image. The weights decide how much of each prompts is incorporated in the final image. Because of this process results of multiprompts are always different than single prompts with the same words in.
As a result of multiple prompts being resolved, fast time can be consumed a little quicker for multiprompts.
3. How to use[edit | edit source]
Its possible to combine many many prompts into a single command but the results will become more and more unpredictable as you do so.
Tip 1: Start with a single prompt first and use multiprompting if you cannot achieve the desired outcome with a single prompt.
Tip 2: Make sure that none of your prompts is a null prompt; a prompt without a subject.
For example:
The prompt 'beautiful' has no subject, which will probably result in the model trying to fit in a subject of its own, such as a portrait of a woman.
Tip 3: For more control, make sure that the prompts you wish to combine have an overlapping element. For example: repeat you would like to have a full body view in multiple/every one of the prompts.