Describe Command

The /describe command allows you to upload an image and generate four possible prompts based on that image. Use the /describe command to explore new vocabulary and aesthetic movements.

PreviousBlend Command NextParameter

Last updated 1 year ago

Describe Command

The /describe command allows you to upload an image and generate four possible prompts based on that image. Use the /describe command to explore new vocabulary and aesthetic movements.

/describe generates prompts that are inspirational and suggestive, it cannot be used to recreate an uploaded image exactly. /describe returns the aspect ratio for uploaded images.

Midjourney’s new “describe” feature is nothing new to the avid prompt crafter. Using tools like or you could already analyze images, get a prompt that would recreate it (or something similar), and take it from there.

Midjourney AI, the leading AI image generation tool has just added another bombshell feature. Now we are able to input images to get text outputs.

Why does it matter?

We can now get a glimpse of how the prompting engine works.
We can recreate images from all over the internet
We can explore new realms of possibilities

Here's how to do it

First, you will need a reference image ready on your computer. This can be a photo, an AI-generated image, or anything else that is saved as an image.

For the sake of this example, I will use a photo from Pexels.

Now let's go to Midjourney and type in the new /describe command.

Once we hit enter we are prompted with an image input form. Let's choose the image we just downloaded.

And hit enter once again.

Now we get four versions of descriptions for the image:

The buttons below allow you to generate new images from these descriptions. Let's see how close to the original these get.

I would say Midjourney did a pretty good job at recreating the original image.

How you can use this?

Once this was released I immediately thought of several ways to use the new feature. Let me share some of these ideas with you.

1. Recreate images/style from anywhere

As with the example above, we can download any image from the internet and attempt to create something similar. Let's try one more. But this time something more abstract.

Here are the descriptions Midjourney came up with:

1️⃣ a purple and white abstract painting painted on canvas, in the style of dark pink and white, detailed backgrounds, clear colors, neo-plasticist, gutai group, paint dripping technique, frenzied action painting --ar 4:3

2️⃣ abstract purple textured background with white paint on it, in the style of dark pink and white, fused abstraction, neo plasticism, crystal cubism, jean-paul riopelle, high quality photo, phyllida barlow --ar 4:3

3️⃣ a close up of the painting with lots of pink, in the style of dark magenta and white, poured paint technique, high quality photo, fragmented planes, vibrant acrylic colors, monochromatic masterpieces, light purple and white --ar 4:3

4️⃣ a close up photo of an abstract and purple piece of glass, in the style of dark pink and white, color zone painting, artistic fragments, pattern-based painting, paint dripping technique, monochrome painting, vik muniz --ar 4:3

And here are the results:

2. Improve your prompting skills

By learning from the descriptions Midjourney generates we can get a glipse of how the language processing part of the tool works. Analyzing these outputs will help us make better prompts in the future.

One thing I already noticed is that Midjourney almost always refferences an artist. And indeed in version 5 using artists as a referrence in your prompts works pretty well.

3. Remix our images in a new way.

We can turn our existing generations into something new by using /describe and then generating new variations of our existing images.

Let's give it a go.

Here's an image I used in one of my previous articles:

And here are some new variations I created using the /describe feature

Here are the prompts:

The original Prompt: beautiful woman actress kissing a bot, full body 

1️⃣ a lady kissing a robot, in the style of golden age aesthetics, datamoshing, emotive body language, precisionist, photo taken with provia, schizowave, golden age aesthetics --ar 25:14

2️⃣ futuristic woman kisses robot on the lips, in the style of romantic depictions of historical events, i can't believe how beautiful this is, photobash, konica auto s3, gray and brown, romantic emotion, iconic album covers --ar 25:14

3️⃣ the kiss between woman and robot in the museum of science and technology at dusseldorf, in the style of joel robison, cinematic elegance, ralph mcquarrie, hendrik kerstens, silver and emerald, animated gifs, stock photo --ar 25:14

4️⃣ robot love for a teenage girl, in the style of peter kemp, cinematic composition, john wilhelm, animated gifs, distinctive noses, kurt wenner, silver and cyan --ar 25:14

4. Altering the results from `/describe` to get something new

Let's combine prompts from two of our previous results to get an image in a different style.

Let's go with:

️⃣ a lady kissing a robot, in the style of golden age aesthetics, datamoshing, emotive body language, precisionist, photo taken with provia, schizowave, golden age aesthetics --ar 25:14

And:

a purple and white abstract painting painted on canvas, in the style of dark pink and white, detailed backgrounds, clear colors, neo-plasticist, gutai group, paint dripping technique, frenzied action painting --ar 4:3

Let's try to capture the style of the abstract image, while maintaining the girl kissing a robot scene.

Combined prompt:

 a lady kissing a robot abstract painting painted on canvas, in the style of dark pink and white, detailed backgrounds, clear colors, neo-plasticist, gutai group, paint dripping technique, frenzied action painting --ar 25:14

And here's the result:

Final words

Once again Midjourney has added another great feature to their already amazing image generator. Can't wait to see what's to come. There has never been a better time to explore your creativity than now.