Update.... I upgraded to source control.... Post is updated with newest version

https://github.com/Midgard-Public/Midgard-Theory-Of-Layer-Separated-Midjourney-Prompting

I spent a few hours running a meticulous experiment with chat gpt 4 to create amazing prompts for Midjourney 5 alpha. Then I spent even more time figuring out how to ask it to adjust images.

Then I asked it to explain the theory we had created in a way that you could paste it into a new chat gpt 4 chat and ask for a list of what you want and immediately get good prompts like I was getting.

This is what came out and it works for me. Copy what is below and paste into ChatGPT 4

Note : This is the version I'm going to keep updatings as the experiment continues.

​

--------------------------------------------------------------------------------------

​

Midgard's Theory of Layer-Separated Prompt Generation For Midjourney With Chat GPT (The Theory):

​

Definitions:

​

Comma-separated sections: Semantically merge ideas within a layer.

Layer separators: Define the balance between the contributions of larger semantic clauses over the entire image.

Guides: Factual statements about Midjourney prompting or important rules to follow.

Tips: Special case rules, helpful rules of thumb, or guidance when choosing between different solutions.

​

Guides:

​

Break the description into multiple layers, focusing on distinct aspects of the subject.

Assign weights to each layer (::X, where X is a number) based on the importance or prominence of that aspect.

Maintain a large dynamic range between the initial clause and any subsequent clauses, allowing the scale to adjust as low or high as necessary.

Use negative weights to negate unwanted subjects or aspects, but ensure no zero or negative total layer weight.

Include specific nouns from the request in the first clause.

Emphasize descriptive nouns and visually descriptive phrases in the prompt.

Utilize terms from relevant fields, such as art techniques, artistic mediums, and artist names, when describing styles.

Stay within a token limit (e.g., 250 tokens) to ensure the entire list can be generated by ChatGPT.

Don't use negative logical words like no, as Midjourney only understands positive phrases.

For any negations, use negative layer weights.

Always try to use the maximum number of tokens in each prompt, emphasizing additional harmonious descriptive styling and especially artists whenever possible.

Have one high weight, one or two middle weight, and many light weight layers.

​

Tips:

​

Layer weights are normalized, so focus on relative values, not absolute values. Always normalize so the lowest non-zero non-negative prompt weight is 1 when adjusting layer weights. Using floating point numbers is acceptable if it helps.

Avoid phrases that don't have a literal visual meaning unless integral to the request, as they can result in unpredictability.

For descriptive styling, use short clauses separated by commas, combining compatible artists and styles when a genre is suggested.

Experiment with adjusting the layer weights until you achieve the desired emphasis in the final result.

Modify prompts by changing layer weights, adding/changing/deleting words and phrases, or rearranging/adding/deleting layer separators.

Employ more common language and concrete terms when modifying prompts for better results.

When using negative layer weights, never negate a term that's also present in the positive part. Negation is most effective when a word has multiple meanings and the correct one must be selected.

​

Example:

​

Original prompt: Create an art piece featuring a serene landscape with a waterfall and a rainbow in the Impressionist style.

​

Scenic landscape, waterfall ::8. Birds flying, rainbow ::4. Impressionist style, Claude Monet ::2. Lush greenery, vibrant colors ::2. Serene atmosphere, calming nature ::1.

​

Let's say we want to emphasize the rainbow and change the art style to Art Nouveau.

​

Adjusted prompt:

​

Scenic landscape, waterfall ::8. Birds flying, rainbow ::6. Art Nouveau style, Alphonse Mucha ::4. Lush greenery, vibrant colors ::2. Serene atmosphere, calming nature ::1.

​

By following The Theory, you can create effective layer-separated prompts for ChatGPT. Use this template for generating prompt lists:

​

Please give me a list of prompts for :