Improve Your Midjourney Art with the /describe Tool

Have you ever seen a great image and wished you knew exactly what prompt was used to create it? Maybe you saw a cool picture of a person in pink and gold armor against a sunny sky. How would you even begin writing a prompt for that?

Instead of guessing, you can simply ask Midjourney what it sees in an image. This guide shows you how to use the built-in /describe feature to get potential prompts directly from a picture. We'll also cover how to make those prompts give you results that are even closer to the original image.

Starting with /describe

The easiest way to start is by using the /describe command in Discord. Make sure the image you want to analyze is saved on your computer.

Here's how it works:

• Type `/describe` in the Midjourney prompt box.

• You'll see an option to upload a file. Click this and select your image.

• Hit Enter.

Midjourney will analyze the image and give you four different prompt options. These prompts are Midjourney's interpretation of the picture you showed it. For our pink and gold example, you might get prompts like "a pink and gold robotic person" or "futuristic female portrait in pink and blue colors."

Below these prompts, you'll see numbered buttons. Click any of these numbers, and Midjourney will generate four images based on that specific prompt. You can try clicking each one to see the different results. Sometimes the initial results are great, but sometimes they aren't quite what you hoped for.

Getting Closer to the Original Image

If the images generated from the /describe prompts don't look enough like your original picture, you can do a few things to guide Midjourney.

Use the Original Image as an Image Prompt

This is a powerful technique. You can use the original image itself as part of your new prompt created by /describe.

Here's the process:

• Click the plus (+) button next to your prompt box in Discord.

• Select "Upload File" and choose your original image.

• Hit Enter to upload it.

• Once uploaded, click on the image in Discord so it expands.

• Right-click on the expanded image and select "Copy Image Address."

• Now, take one of the prompts you got from the /describe command.

• Paste the image address you copied * before * the text of the /describe prompt. It will look something like: `[image address] [describe prompt text]`

• Hit Enter to generate.

Adding the original image link gives Midjourney a visual reference point, helping the new generations look much more similar to the source image. This is incredibly helpful for getting accurate results.

Control Image Influence with --iw (Image Weight)

You can tell Midjourney how important that image prompt link is compared to the text prompt. Use the `--iw` parameter at the end of your prompt.

• The default image weight is typically 1, but you can set it from 0.5 up to 2.

• `--iw 0.5` means the image prompt has less influence than the text.

• `--iw 2` means the image prompt has twice the influence of the text.

Experimenting with image weight can make a big difference in how closely the generated image matches the style, composition, or subject of your reference picture. For example, increasing the weight to 2 might result in more images facing the same direction as your reference.

Managing multiple image prompts and testing different weights can get repetitive quickly. Consider using a tool designed to handle these steps for you, like the Midjourney Automation Suite from TitanXT, which can streamline experimentation.

Adjust Stylize and Chaos

Even with an image prompt, you might want to fine-tune the variety and artistic interpretation. The `--stylize` and `--chaos` parameters help with this. You can use these with or without an image prompt link.

• --stylize (or --s): Controls how much artistic license Midjourney takes. Higher values (up to 1000) mean Midjourney adds more artistic flair, potentially moving further from your exact prompt. Lower values (towards 0) make Midjourney stick more strictly to your words, but can sometimes look less polished if the prompt isn't perfect. A value around 400 often gives good results.

• --chaos (or --c): Influences the variety within the grid of four images Midjourney generates. A value of 0 (default) means the four images will be quite similar. Increasing the value up to 100 makes the pictures in the grid look more different from each other.

Combining these parameters gives you more control over the final output when using prompts derived from /describe.

Other Useful Tips for /describe

Here are a few extra things to keep in mind when using /describe:

• Re-roll /describe: If the initial four prompts given by /describe aren't quite right, you can click the re-roll button below them to get four new prompt suggestions.

• Artist Names: Sometimes /describe will include an artist's name. Be aware that if the name isn't hyperlinked, it might be a made-up name that Midjourney can sometimes still interpret.

• Aspect Ratios: The aspect ratios /describe outputs might look slightly different than standard ratios (like 91:51 instead of 16:9). This is because Midjourney often rounds to the nearest 32-pixel value during generation, so even if you prompt 16:9, the output might not be exactly that ratio when then analyzed by /describe.

• Text in Images: While /describe can sometimes read text within an image (like "find your balance" on a picture of a lake), Midjourney itself currently cannot reliably generate text based on the description. Just because /describe sees it doesn't mean Midjourney can recreate it.

Conclusion

Midjourney's /describe feature is a fantastic starting point for generating prompts based on existing images. It provides a solid foundation, but you can get even better results by combining the generated prompts with an image prompt of the original picture, adjusting image weight, and tweaking stylize and chaos values.

Refining prompts and parameters takes time and experimentation. Tools like the Midjourney Automation Suite from TitanXT can help automate the process of testing variations derived from your /describe results, saving you time and effort.

Give /describe a try and see how it helps you create more accurate and desired images!