Midjourney Image Prompts: Using Image URLs and --iw Properly

What an image prompt is

An image prompt is a picture you give Midjourney as part of the prompt itself. Midjourney looks at the whole image, including its subject, layout, colours and overall feel, and blends that with your text. It is not an editor: it will not keep your exact photo and change one detail. It treats the image as strong inspiration.

The structure is always the same:

1. One or more image URLs first.
2. Then your text prompt.
3. Then your parameters, each starting with --.

For example:

https://example.com/my-sketch.png a cosy reading nook under a staircase, warm lamp light, children's book illustration --ar 3:4 --iw 1.5

On the Midjourney website you can also drag or upload an image into the prompt bar and mark it as an image prompt, which inserts it for you. Whichever way you add it, the image must be publicly reachable or uploaded to Midjourney, and should be a direct link to the image file rather than a web page.

Controlling influence with --iw

--iw sets how much the image prompt matters relative to your text. Higher values make results look more like the image. Lower values give your words more control. The allowed range and default have changed between model versions, so check Midjourney’s docs for the current numbers, but the behaviour is consistent:

A practical approach is to run the same prompt at a low, default and high weight and pick the one that balances your reference and your idea.

Remember that image weight works alongside your other settings. A high --stylize value or a strong style reference can pull results away from the image even at a high --iw, so change one thing at a time when you are testing.

Image prompts vs style references

This is the most common point of confusion. Both take an image, but they do different jobs:

• Image prompt (URL at the start, --iw) influences what is in the picture: subject, composition, shapes and colour as well as style.
• Style reference (--sref, --sw) borrows the look only: palette, texture, rendering and mood. The subject comes from your text.

If you use a photo of a red armchair as an image prompt, you will probably get armchairs. Use it as a style reference and ask for "a lighthouse", and you will get a lighthouse with that photo’s colours and light. You can also use both in one prompt: an image prompt for composition and an --sref for the look. Our style reference guide covers --sref in depth.

A practical workflow

1. Decide what the image is for. Composition, subject, palette or all three? If it is only palette and mood, switch to --sref.
2. Prepare the image. Crop out distractions. A clean reference gives a cleaner result. Match its aspect ratio to the one you plan to use where you can.
3. Write text that adds, not repeats. The image already shows what it shows. Use your words for the changes: new setting, time of day, medium or mood.
4. Run a weight test. Same prompt at low, default and high --iw. Fix a seed if you want the comparison to be fairer.
5. Refine with Vary. Once a result is close, use Vary (Subtle) for small changes or Vary (Strong) for bigger ones rather than starting again.
6. Save what worked. Keep the winning image URL, text and weight together so you can reuse the recipe.

Copy-ready examples (replace the URLs with your own):

• https://example.com/floorplan-sketch.jpg bright Scandinavian kitchen, oak cabinets, white walls, morning light, architectural photography --ar 16:9 --iw 2
• https://example.com/bottle.jpg perfume bottle on wet black stone, moody studio lighting, water droplets --ar 4:5 --style raw
• https://example.com/forest.jpg https://example.com/city.jpg a city reclaimed by forest, soft fog, golden hour --ar 21:9

Tips and common problems

• The result ignores my text. Lower --iw, or make your text more specific about what should change.
• The result ignores my image. Raise --iw, and remove text that contradicts the image.
• It copied the style but I wanted the layout. That is an image prompt doing its whole job. Add a different --sref to override the style, or describe the style you want.
• I want the same person in every image. Image prompts help but rarely hold a face perfectly. See our guide to consistent characters.
• Text or logos come out garbled. Midjourney is unreliable at reproducing long exact text, even from an image. Add precise typography and logos afterwards in a design tool.
• Rights. Only use images you own or have permission to use as references.

Doing this faster with Dreamdrive

Dreamdrive saves every prompt you run automatically, with its picture, model and settings, into the project you choose, so you can search your Library for it later, copy it with Copy prompt or drag the picture into another tool.