Get the Prompt of any Image with Midjourney

With Easter upon us, the engineers at Midjourney have been delivering some amazing Easter eggs for us to enjoy over these holidays. There are two BIG features that are /describe and permutations that are next level in AI text to image creation game.

In this post we will look at the /describe feature which an image to text generation, in other words take any image and upload it to have Midjourney describe it back to you in words. It let’s you decipher or reverse engineer any image to get its prompt that AI thinks it could be. Take those prompts as is or a good starting point and use /imagine to create many more similar images.

Of course you will not get 100% exactly same image but you will get quite close and you can always modify the prompt to add your own touch to it.

The /describe function is similar to the CLIP interrogator that takes and image and creates the text based prompt that represents the image. I have done earlier posts on this topic: Clip Interrogator and there a second version available via the Tools page.

How to use /describe

For this experiment I take the photograph by Duncan Lewis on Unsplash

In your discord, type the command /describe and an upload button will appear. Upload any desired image, can be AI generated image or real photograph. Midjourney will then process the image and create 4 different prompts that describe the image.

Here are the prompts that Midjourney created for our selected image and you also get the aspect ratio of the uploaded image:

1️⃣ colorful street with neonlighted buildings, in the style of japanese photography, futurist claims, light magenta and dark emerald, hip-hop culture exploration, japonism influenced pieces, chaotic academia, kimoicore –ar 3:4

2️⃣ a street in tokyo at night, in the style of light emerald and magenta, hip hop aesthetics, science-fiction lands, color splash, oku art, maximalism, asian-inspired –ar 3:4

3️⃣ a pedestrian walkway in an japanese neighborhood, in the style of neon hallucinations, cyberpunk, light emerald and magenta, old-world charm, urbancore, nightscape, onii kei –ar 3:4

4️⃣ an illuminated city nightscape tsukuba jinga, japan, in the style of street scenes with vibrant colors, light emerald and magenta, queer academia, matte photo, street scene, futurism influence, eye-catching tags –ar 3:4

Let’s now take all four of these prompts and create new images from them.

Image 1 – Colorful Street

Image 2 – A street in Tokyo

Image 3 – A pedestrian walkway

Image 4 – An illuminated city nightscape

Review of /describe

Reviewing the images generated by the prompts created using /describe, Image #3 is the most closest to the original photograph selected. Others are representing more of a street scene with roads and cars, while the original photograph is from a laneway which is paved and a few people roaming the area.

As I stated earlier in the post the /describe generated prompt is not exact and only attempts to start to describe the image, you should always review and add missing elements and remove any words that are emphasising unwanted results in the final image. So reviewing carefully at the Image #3 prompt, there are several things I’d like to fine-tune: a pedestrian walkway in an japanese neighborhood, in the style of neon hallucinations, cyberpunk, light emerald and magenta, old-world charm, urbancore, nightscape, onii kei –ar 3:4

I want to add that its paved, has neon signs and lights and remove the terms like neon hallucinations, cyberpunk and the colour emphasis by emerald and magenta. So my refined prompt is now: a paved pedestrian walkway in an japanese neighborhood, neon signs, neon lights, old-world charm, urbancore, nightscape, onii kei, couple of people roaming –ar 3:4 –v 5

Now with this refined prompt I am able to get following images

Now I think we are very close to the original image that was taken from Unsplash and have been able to create similar looking images using Midjourney. Have a play with the /describe feature take your favourite photograph or AI generated image and see how Midjourney sees the image.

If you'd like to support our site please consider buying us a Ko-fi, grab a product or subscribe. Need a faster GPU, get access to fastest GPUs for less than $1 per hour with RunPod.io