Midjourney /describe Image-to-Text Prompt Generator
What words do I use in my prompt? How do I know what style I want?
These are common questions many of us have when first using an AI image generator like MidJourney. Even those of us with months and years of experience still ask these questions. AI image generator algorithms are constantly changing, and MidJourney is no exception.
Having a collection of words to pull from for a specific aesthetic can be invaluable as you write text prompts. In my opinion, this is one of the greatest benefits to using image-to-text generators.
MidJourney's /describe command is an image-to-text feature that will give you sample prompt suggestions. You can immediately create images with these suggestions through /imagine, or you can take specific words from them and create new prompts.
In this article, I'll show you how to use MidJourney /describe and some examples of it in action!
How to use MidJourney /describe image-to-text
Diving into the results
Rolling the text prompt suggestions
More image-to-text prompt examples with MidJourney /describe
Summary
How to use MidJourney /describe image-to-text
Using MidJourney's /describe image-to-text feature is easy and straight-forward. In Discord, you simply type /describe, drag and drop your image, and hit the Enter key. Note: To utilize the /describe image-to-text feature, you must have an active MidJourney membership. If you're looking for a free image-to-text generator, check out this article.
In Discord, type /describe instead of the usual /imagine command and then add a space
Drag and drop your image file into the window or click to upload, then press Enter
In seconds, you'll receive 4 prompt suggestions from MidJourney based on your uploaded image
Click one of the numbered buttons below your image to instantly submit an /imagine command for the prompt, or click "Imagine all" to submit all four prompts
To generate new text prompts on the same image, simply click the re-roll button
Submitting your image with the /describe command
Diving into the results
First thing's first -- the algorithm used by /describe is NOT the same as what is used by MidJourney's /imagine command.
What does this mean?
Prompt suggestions given by /describe may need modification. Depending on the style and contents of the image, some prompt suggestions will need more tweaking than others. MidJourney /describe provides prompt suggestions, not exact prompts that will allow you to perfectly duplicate your input image. The prompt suggestions are guides to help you write prompts and gain creative insight.
Ok, let's get into the results for the sci-fi-esque image I submitted to /describe.
MidJourney's /describe command gave me four prompt suggestions for my input image:
image: a female in a gas mask standing in an underwater situation, in the style of dark orange and aquamarine, photo-realistic landscapes, zbrush, packed with hidden details, orderly symmetry, transportcore, futuristic landscapes --ar 3:2
a woman in oxygen mask standing in the ocean, in the style of hyper-realistic sci-fi, dark orange, photorealistic, urban scenes, uhd image, transportcore --ar 3:2
marine woman underwater in a trenchcoat surrounded by water and starfish, in the style of futuristic cyberpunk, orange, frostpunk, 8k 3d, packed with hidden details, industrial technological subjects, traincore --ar 3:2
a man who has a gas mask in a forest scene with bubbles, in the style of dark orange aquamarine, hyper-realistic sci-fi, detailed crowd scenes, daz3d, intricate underwater worlds, transportcore, passage --ar 3:2
Across these four prompt suggestions, you can see themes emerge that fit the style of the input image: "dark orange and aquamarine", "gas mask", "underwater", "hyper-realistic sci-fi". We also have fun, odd things like "frostpunk" and "transportcore".
At the end of each prompt, MidJourney has included --ar 3:2. This means that it recognized the aspect ratio of our input image as 3:2, so it has included it in all four prompt suggestions.
Rolling the text prompt suggestions
I submitted /imagine commands for all four of my prompt suggestions by clicking the "Imagine all" button. Below are the image grids I got back for each of the prompts. A couple of the prompts got pretty close to the original aesthetic of the orange and aquamarine sci-fi underwater-esque look.
/imagine results from prompt suggestion #1
/imagine results from prompt suggestion #2
/imagine results from prompt suggestion #3
/imagine results from prompt suggestion #4
More image-to-text prompt examples with MidJourney /describe
Now that you've seen how the /describe command works and some example prompt results, let's take a look at a couple more examples.
In the next example, I uploaded a picture (created with MidJourney) of a very cute dog resting on a rock above some Norwegian scenery. A couple of the /describe prompt suggestions immediately picked up on the Norwegian scenery. Again, MidJourney picked up that the input image had an aspect ratio of 3:2, so it included --ar 3:2 at the end of each prompt suggestion.
I rolled all four prompt suggestions to see if I could get a similar image and style to my input image. The /imagine results from prompt suggestions #2 and #3 looked most similar to my input image. My favorite results though were from prompt suggestion #4.
/imagine results from prompt suggestion #1
/imagine results from prompt suggestion #2
/imagine results from prompt suggestion #3
/imagine results from prompt suggestion #4
For this last example, I uploaded a Scandinavian city inspired image that I made with Midjourney. All of the /describe prompt suggestions included references to the buildings, skyline, and water. Three of them referenced contemporary Scandinavian art.
I rolled all four prompt suggestions. The color palettes across all four results grids were similar to my input image with the colorful buildings. Results from prompt suggestion #1 share the most similarity with my original image.
/imagine results from prompt suggestion #1
/imagine results from prompt suggestion #2
/imagine results from prompt suggestion #3
/imagine results from prompt suggestion #4
Summary
Midjourney's /describe image-to-text prompt generator is a great tool to use for quickly creating text prompts. Simply upload an image that has an aesthetic or theme that you like and let MidJourney /describe give you prompt suggestions.
The prompt suggestions can be especially useful when you are looking for unique ways to describe the contents of an image. The suggestions can help you get past creative blocks and provide a starting point for new prompts.
MidJourney /describe does the best it can with your input image and the algorithm it uses. The results will not always be accurate or make sense, but for the most part, I find its suggestions a great starting point for writing new prompts.
Want to use a free image-to-text generator? Check out the article below!
https://woollyferncreative.com/blog/2023/07/23/image-to-text-prompt-generators-midjourney-describe-clip-interrogator-llava/
Thanks for reading!