How do I use image prompts, Describe, and Blend in Midjourney?

All three features control generations with images instead of words alone: an image prompt steers the result toward your reference, Describe turns an image into prompt text, and Blend merges several images into one.

Image prompts (image-to-image)

An image prompt is a reference picture placed before your text prompt; the result borrows its composition, colors, or character.

Web app:

1. On the Create page, click the image icon (+) on the left side of the prompt box and upload or pick an image.
2. A thumbnail appears in the box. Choose how it's used: Image Prompt (overall reference), Style Reference (style only), or Character Reference (character only).
3. Add your text prompt as usual and submit.

Discord:

1. Send your reference image to the bot, open it, and copy the image link.
2. In /imagine, paste the link at the very start of the prompt, then add your text:

https://.../ref.png a warrior in this style, full body --ar 2:3

1. To make the reference count for more, add --iw 2 (image weight, 0–3, default 1).

Describe: image → prompt

Want to know how a great image "should be written"? Use Describe:

• Web app: upload the image, then choose Describe from its options.
• Discord: run /describe and attach the image.

Midjourney returns 4 candidate prompts. Generate directly from one, or lift the style keywords into your own prompt — this is the fastest way to learn prompt writing.

Blend: merging images

Blend fuses 2–6 images with no text needed:

• Discord: type /blend and upload two images (click the optional fields for up to 6). The dimensions option sets the output ratio.
• Web app: upload several images together as image prompts — same effect.

Great for "character from image A in the art style of image B." For finer control, follow a Blend with Vary to refine the result.

Troubleshooting

• Result strays too far from the reference: raise --iw, or switch to a Style/Character Reference for targeted borrowing.
• Upload rejected: the image itself tripped content moderation — use a different reference.
• Describe output doesn't recreate the original: Describe describes, it doesn't reconstruct. To get close to the original, combine an image prompt with the Describe text.