Writing Prompts: Structure, Multi-Prompts, and Image Inputs
- Structure a prompt with subject, style, lighting, and composition to get predictable results
- Use the :: separator to weight different concepts within a prompt
- Add image URLs as inputs and control their influence with --iw
The Anatomy of a Good Prompt
Midjourney prompts work best when they describe the image you want across a few key dimensions. Think of each prompt as answering these questions:
- Subject: What is in the image? A person, object, scene, creature?
- Style / Medium: Is it a photograph, oil painting, pencil sketch, 3D render?
- Lighting: Golden hour, studio lighting, overcast, neon, candlelight?
- Composition: Close-up portrait, wide establishing shot, overhead view?
- Color / Mood: Warm tones, desaturated, vibrant, noir?
You don't need all five in every prompt. But the more relevant context you give, the less Midjourney has to guess. a cat will produce something. a tabby cat sitting on a windowsill, late afternoon light, shallow depth of field, film photograph will produce something much closer to a specific vision.
What Works — and What Doesn't
Midjourney responds to adjectives and visual descriptors far better than instructions. It understands what things look like, not what actions to take.
- Works:
a weathered leather journal on a wooden desk, warm lamp light, macro photography - Doesn't work:
make the background blurry and the subject sharp— this is an instruction, not a description. Sayshallow depth of fieldinstead. - Works:
bold geometric shapes, primary colors, Bauhaus design poster - Doesn't work:
design a poster that uses bold shapes— Midjourney isn't reading instructions, it's interpreting descriptions.
Specificity almost always helps. If you want a forest, say whether it's a redwood forest, a misty Scottish pine forest, or a Japanese bamboo grove. Each will produce dramatically different images.
Multi-Prompt Weighting
The :: separator lets you split a prompt into parts and assign relative importance to each part. Without it, Midjourney treats your entire prompt as a single idea. With it, you can tell it to focus more on one concept than another.
Syntax: concept one ::2 concept two ::1
The numbers are weights. A weight of 2 means that concept gets twice as much influence as a weight of 1. Negative weights remove elements: forest ::2 snow ::-1 asks for a forest and actively discourages snow.
A practical example: hot dog gives you a frankfurter. hot ::2 dog ::1 gives you a warm, sweaty dog in summer heat — because the separator breaks the compound word into two concepts.
Use multi-prompt weighting when you want precise control over which elements dominate the composition, or when a phrase is being misinterpreted as a compound word.
Using Images as Inputs
You can use an image URL as the start of a prompt to give Midjourney a visual reference. Paste the URL at the very beginning of your prompt, before any text.
Example: https://example.com/photo.jpg a portrait in the style of that photo, studio lighting
The --iw parameter (image weight) controls how much influence the input image has over the composition. Values range from 0 to 3; the default is typically around 1. Higher values make the output look more like the input image.
--iw 0.5: light influence — the text prompt drives most of the image--iw 1: balanced — roughly equal influence from image and text--iw 2: heavy influence — the output will closely resemble the input image
Image inputs are useful for maintaining composition from a reference photo, recreating the mood of an existing image, or using your own sketch as a starting point. For maintaining specific characters or objects across multiple generations, the dedicated Omni Reference tool (Lesson 7) is more powerful.
Negative Prompts
The --no parameter tells Midjourney what to leave out. It goes at the end of your prompt.
a field of sunflowers, summer afternoon --no people
You can stack multiple items: --no people, text, watermarks. This is useful for removing common distractions — Midjourney sometimes adds text or logos to images unprompted, and --no text, watermarks, logos suppresses that tendency.
- Midjourney responds to visual descriptors, not instructions — describe what the image looks like, not what to do
- The :: separator lets you weight concepts (hot ::2 dog ::1 gives a warm dog, not a frankfurter)
- --iw controls how much an input image URL drives the output — 0 to 3, higher = closer match
Progress is saved in this browser only — clear your cache or switch device and it's gone. Create a free account and it follows you everywhere.