Verified against Midjourney · 2026-08-07
Remix an uploaded reference photo with an image prompt and --iw
A foundational image-prompt workflow — pairing an uploaded reference image URL with text instructions and a tuned --iw weight — for restyling a real photo, turning a rough sketch into finished art, or blending a real scene's composition with new stylistic direction, without touching --cref, --sref, or --oref's more specialized, narrower mechanisms.
The prompt
Ready to copy — highlighted parts are example details you can swap.
REFERENCE IMAGE https://cdn.midjourney.com/mno345-rough-sketch.png — the uploaded photo, sketch, or existing image whose composition and general content this generation should draw from. WHAT IN THE REFERENCE SHOULD CARRY THROUGH the overall composition and camera angle, and the rough placement of the three main figures — name specifically which parts of the reference matter most (the composition and camera angle, the color palette, the general layout of objects), since --iw weights the whole reference image as one signal and cannot selectively preserve one element while ignoring another on its own — the text prompt has to do that selective work. TEXT INSTRUCTIONS FOR THE NEW RESULT rendered as a finished, fully painted digital illustration in a warm watercolor style, with detailed lighting and color the rough sketch lacks — what should change or be added relative to the reference: a new style, a different medium, additional detail the reference didn't have. IMAGE WEIGHT --iw 1.5 — this sets how strongly the reference image's actual visual content competes against the text instructions. Low values (0.5-1) let the text direction dominate, using the reference mostly as loose compositional inspiration. High values (2-3) keep the result much closer to the reference's actual content, using the text mainly to adjust style or add detail on top of it rather than to substantially reinterpret the scene. WHY THIS IS DIFFERENT FROM CREF, SREF, OR OREF This is the general-purpose, original image-prompting mechanism — an uploaded image plus text, weighted against each other — rather than one of the specialized reference types built for a narrower job: --cref specifically for a character's face and clothing, --sref specifically for color and lighting style transferred onto new content, --oref specifically for a single object's exact geometry. Use plain image-prompt weighting when the goal doesn't fit neatly into any of those three narrower boxes — restyling a whole photo, or turning a sketch into finished art — rather than forcing one of the specialized types to do a job it wasn't built for. WHAT TO AVOID --no text, watermark, visible pencil sketch lines PARAMETERS --iw 1.5 --no text, watermark, visible pencil sketch lines --ar 16:9 --v 7 OUTPUT https://cdn.midjourney.com/mno345-rough-sketch.png rendered as a finished, fully painted digital illustration in a warm watercolor style, with detailed lighting and color the rough sketch lacks --iw 1.5 --no text, watermark, visible pencil sketch lines --ar 16:9 --v 7 TUNING --IW ACROSS ATTEMPTS If the result ignores the reference almost entirely and just generates from the text alone, --iw was too low for how much of the reference actually mattered — raise it. If the result barely changes anything from the reference despite clear text instructions to change the style or medium, --iw was too high for what was being asked — lower it, since a very strong image weight can effectively drown out text instructions that are asking for a substantial reinterpretation rather than a light touch-up.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Plain image-prompt weighting is the original, general-purpose version of what --cref, --sref, and --oref later specialized into narrower jobs — all four mechanisms ultimately weight an uploaded image's visual content against the text prompt's instructions, but the three specialized reference types constrain what part of the image gets matched (a face and clothing for --cref, color and light for --sref, one object's geometry for --oref), while plain --iw weights the entire reference image as one undifferentiated signal. That difference is exactly why plain image-prompt weighting is the right tool for jobs that don't fit neatly into one of those three narrower boxes — restyling a whole photograph's medium, or turning a rough sketch into a finished painting — where the goal is closer to "use this whole image as a loose or strong starting point" than "match this one specific attribute of it."
Requiring an explicit statement of which elements of the reference matter most, rather than assuming --iw handles that selectivity on its own, addresses a real limitation of the parameter: it adjusts one overall dial for how strongly the reference competes against the text, but it has no internal mechanism for preserving the reference's composition while discarding its color palette, or vice versa — any selective preservation has to come from what the text prompt explicitly asks for, since the image weight alone only controls how much of the reference bleeds through as an undifferentiated whole, not which specific parts of it do.
The distinction between low --iw ("loose inspiration, text direction dominates") and high --iw ("close to the reference, text mainly adjusts style on top") reflects the actual mechanical trade the parameter makes between two competing signals feeding into one generation — it is not a linear "more reference equals better fidelity" dial, it is a balance point, and picking the wrong side of that balance for the actual goal produces two specific, recognizable failure modes: a too-low setting on a job that needed the reference's composition preserved produces a result that might as well have skipped uploading the reference entirely, while a too-high setting on a job that wanted a substantial reinterpretation produces a result barely different from the source despite clear text instructions asking for real change.
Naming both of those specific failure signatures, and mapping each to the actual direction to adjust --iw next, turns a discouraging "the remix didn't work" result into an immediately actionable next step, rather than a rewritten text prompt that leaves the actual mismatched parameter untouched and produces the same kind of disappointing result on the next attempt.
Verified against
Midjourney v7 · 2026-08-07
Changelog
- 2026-08-07 — Initial publish, verified against Midjourney v7 image-prompt --iw weighting for photo and sketch remixing.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
