Verified against Midjourney · 2026-07-29
Carry one specific object into different scenes with --oref
An omni-reference workflow using Midjourney V7's --oref/--ow parameters to keep a specific physical object — a product, a prop, a mascot — visually identical across a set of different environments, without --cref's face-and-clothing bias or --sref's style-only scope.
The prompt
Ready to copy — highlighted parts are example details you can swap.
OBJECT REFERENCE One clear image of the specific object that must stay visually identical across every generated scene — the exact product, prop, or mascot, ideally photographed or rendered plainly with nothing else in frame. Its URL goes in https://cdn.midjourney.com/ghi789-product-reference.png. OBJECT DESCRIPTION a matte-black ceramic pour-over coffee dripper with a walnut wood collar and a small embossed logo on the base — restate this in every scene's prompt even though --oref is carrying the visual reference, since the text description still anchors what the model believes it is looking at. SCENE THIS OBJECT APPEARS IN resting on a rustic wooden picnic table at a farmers market, morning sun and blurred stalls behind it WHY THIS OBJECT NEEDS TO STAY IDENTICAL this is a real product for a hero-image set across five different lifestyle placements, and any shape drift would misrepresent what a customer actually receives OBJECT WEIGHT --ow 250 — this controls how strongly --oref's visual reference overrides the model's own interpretation of the object as described in text. Start around 100-300 for a firm match; if the object has fine, specific detail that keeps getting smoothed over or reinterpreted (a specific label design, an unusual shape), push --ow higher rather than adding more adjectives to the text description, since the reference image already contains that detail more precisely than words can. DIFFERENCE FROM CHARACTER AND STYLE REFERENCE --oref exists specifically because --cref is tuned for faces and biases toward matching clothing along with the face, and --sref transfers a color-and-light style but ignores object shape entirely — neither is right for holding a specific inanimate object's exact geometry and surface detail steady while the scene, lighting, and framing around it change freely from shot to shot. WHAT TO AVOID --no text, watermark, a second unit of the product, hands touching it PARAMETERS --oref https://cdn.midjourney.com/ghi789-product-reference.png --ow 250 --ar 4:5 --v 7 OUTPUT a matte-black ceramic pour-over coffee dripper with a walnut wood collar and a small embossed logo on the base in resting on a rustic wooden picnic table at a farmers market, morning sun and blurred stalls behind it --oref https://cdn.midjourney.com/ghi789-product-reference.png --ow 250 --no text, watermark, a second unit of the product, hands touching it --ar 4:5 --v 7 REPEAT FOR THE FULL SET OF SCENES Reuse the identical https://cdn.midjourney.com/ghi789-product-reference.png across every scene in the set, and check each new result specifically for shape and proportion drift on the object itself before checking anything else about the scene around it — a lighting mismatch is easy to fix later in editing, but a subtly wrong object shape means the reference weight or reference image needs adjustment before generating the rest of the set.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
--oref was added in V7 specifically to close a gap the two earlier reference types left open: --cref's matching logic is tuned around faces and, at its default weight, pulls clothing along with the face, which is the wrong bias for an inanimate object with no face at all — and --sref transfers color grading and lighting texture while explicitly not caring about object shape, so it will happily give five scenes the same warm color grade while rendering five subtly different bottle shapes. Neither reference type was built to answer the actual question a product or prop set needs answered — "keep this exact geometry and surface detail identical while everything else in the frame changes freely" — which is precisely the narrow job --oref does and the other two do not. Restating the object's text description in every scene even though the reference image is doing the visual work matters because --oref, like the other reference types, blends a visual signal with the text prompt rather than replacing the text prompt outright; if the text description drifts or gets vague across scenes ("a coffee dripper" in one prompt, "a ceramic pourover thing" in the next), the model has two competing signals about what it's looking at instead of one reinforcing pair, and inconsistency creeps back in even with an identical reference image and --ow value. --ow's role in fixing fine detail — a specific label design, an unusual proportion — rather than adding more descriptive adjectives to the text addresses a real limitation of language as a specification format: a reference image already contains the exact curve of a handle or the precise placement of a logo far more precisely than any string of adjectives could describe it, so when a detail keeps getting smoothed over or subtly reinterpreted across generations, the fix is turning up how much the model trusts the image it already has, not writing a longer paragraph trying to out-describe a photograph. The instruction to check for shape and proportion drift on the object first, before evaluating the rest of the scene, reflects where the actual risk concentrates in this workflow — lighting and background composition vary naturally and acceptably from shot to shot by design, since that variation is the entire point of placing one consistent object into different scenes, but any drift in the object's own geometry is the one failure that defeats the reason --oref was used in the first place, and it is easy to miss if a reviewer's attention goes first to the more visually obvious background instead.
What you get back
A set of scenes where the coffee dripper keeps its exact shape, proportion, and logo placement across a market stall, a kitchen counter, and a studio shot, with lighting and background varying naturally scene to scene as intended.
Verified against
Midjourney v7 · 2026-07-29
Changelog
- 2026-07-29 — Initial publish, verified against Midjourney v7 --oref/--ow for product-object consistency across scenes.
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
