2.3Image prompts & image weight (--iw)
The essentials in 30 seconds
- Show the base: image URL at the start of the prompt — inherited composition, colors and subject.
- --iw doses the influence: low = text dominates, high = image weighs in — balance, don't crush.
- Rights first: your images or free sources — never a protected image as raw material.
Running mini-case — Sonia keeps her room, changes its season
Sonia photographs the real room of the Algiers café and places her photo URL at the start of the prompt: the composition is kept. Then she steers with words ('…but in winter, at dusk, garlands') and doses with --iw: low value when words must dominate, higher when the photo must weigh in. She works on HER photo, not an image found online. In 2.4, this same photo will become a style reference for the series; in 2.5, version and settings will be logged to reproduce.
Sequel to thread 2.2: Sonia's photo sets the base, words steer — and continues in 2.4 and 2.5.
1. Providing an input image
Beyond words, you give Midjourney an input image: an image prompt. Place an image URL at the start of the prompt (on Discord, paste the image then copy its link; on the web, add it as a reference). The model takes inspiration from it for composition, colors, subject. This differs from the style reference (section 2.4): here the image influences content and structure, not only the aesthetic signature — so Sonia keeps her client's real room.
2. Dosing with --iw and combining text + image
The --iw (image weight) parameter sets the image's influence against text: low value = text dominates, high value = image weighs more. Combine image + text: the image gives the base, words steer ('…but in winter, at dusk'). Sonia tries low then high and compares the balance. --iw ranges change with versions: she checks the up-to-date documentation instead of reusing an inherited value.
3. Rights caution
Starting from a protected image (artwork, someone else's photo, brand visual) to generate a derivative raises rights questions — especially for commercial use (see section 0.4, which is not legal advice and varies by country). Sonia's good practices: her own photos first, otherwise free sources with verified license; never deliberately imitate a living artist's style nor reuse a protected element. The image prompt is powerful — use it on raw material whose rights you control.
🔎 Concrete example — 🔎 Worked example — Starting from an image, varying with words
Sonia's café room, kept in structure, changed in mood:
[photo URL of the room] cozy interior, morning light, green plants --iw 1 --ar 4:5
→ keeps the room composition, changes the style.
(lower --iw to let words dominate)Reading: a generation guided by a concrete visual base, not by words alone.
Lesson: adjust --iw to balance image and text, on an image whose rights you hold.
Section vocabulary
- Image prompt
- Image provided as input (via its URL, at the start of the prompt) that Midjourney uses for composition, colors and subject.
Simple example: Sonia places her room photo URL at the head of the prompt.
What it is not: a style reference that conveys only aesthetics.
Reflex: concrete base = URL at the head.
- --iw
- Weight of the input image against text: low = text dominates, high = image weighs more.
Simple example: Sonia lowers --iw so 'in winter' takes over.
What it is not: a format or version setting.
Reflex: balance = try low then high.
- Image + text
- Combine both: the image gives a concrete base (structure, subject), words steer the rest (style, context).
Simple example: The room kept, the season changed by words.
What it is not: an image alone with no written intent.
Reflex: shown base + written intent.
- Rights caution
- Starting from a protected image to derive another raises rights questions; prefer your images or free sources (0.4).
Simple example: Sonia works on her own photo of the café.
What it is not: any image found online as a commercial base.
Reflex: base = controlled rights.
In practice — Hands on — Guiding a generation with an image
- Choose an image whose rights you hold (your photo, a free source) and get its URL.
- Place that URL at the start of the prompt, add a description, and generate: watch what the image brings (composition, colors, subject).
- Vary --iw (low then high) and compare the balance between image and text influence.
- Combine: keep the image base and change the context with words (season, light, style).
Ready-to-use script — explaining --iw dosing to the client (30s)
'For your room, I started from YOUR photo — not an image found online. The URL at the start of the prompt keeps the composition, and words change the mood: winter, dusk. With --iw I dose: low when text must dominate, high when the photo must weigh in. We keep the real room, we explore moods.'
Say while showing the source photo, then the low- and high---iw variants.
Copy-paste template — image base + text intent
[URL of your photo] [kept subject], [new context], [light] --iw 1 --ar 4:5
[text variant] same base low --iw + 'in winter, at dusk' → words dominate
[image variant] same base high --iw → the photo weighs more
[reminder] base = your images or free sources (see 0.4)Room in winter
The café photo + 'in winter, garlands': Sonia keeps the walls, changes the season.
Storefront in the morning
Storefront photo as base, 'morning light' in words: --iw adjusted to keep the sign.
Drinks menu
Photographed cup as base, background changed by words: the cup stays recognizable.
Forbidden material
Someone else's photo found online: Sonia drops it as a commercial base and reshoots her own.
✓ Best practices
- Start from an image whose rights you hold (yours or free).
- Dose --iw to balance image and text influence.
- Combine visual base (image) and intent (words) instead of opposing them.
✗ Common pitfalls
- Deriving a protected image without the right (0.4).
- Confusing image prompt (content/structure) with --sref style reference (2.4).
- Setting --iw to maximum and losing the text contribution.
Key takeaways
- An IMAGE PROMPT provides an input image (URL at the start of the prompt): it influences composition, colors and subject.
- --iw doses the image influence against text: low = dominant text, high = dominant image.
- Combine image + text: the image sets the base, words steer (style, context, variations).
- Rights caution: prefer your own images or free sources; --iw ranges change with versions.
48h challenge — apply tomorrow
With a photo of yours: place its URL at the head, add a worded intent, try low then high --iw, and produce 2 context variants (season, light) while keeping the base.
- I can place an image URL at the start of a prompt.
- I can dose --iw (low = text, high = image).
- I work on images with controlled rights.
Frequently asked questions
What is the difference between an image prompt and a style reference (--sref)?
The image prompt (head URL, dosed by --iw) influences CONTENT and structure — composition, subject, colors. --sref (section 2.4) conveys the aesthetic SIGNATURE — palette, texture, light — to any subject. You can combine both: an image base + a style reference.
Can I use any image found online as an image prompt?
Technically yes, legally discouraged: starting from a protected image for commercial use raises intellectual property questions (0.4, varies by country). Use your own images or free sources with verified license.
What does the --iw parameter set in an image prompt?