Midjourney V8.2 uses a current-generation editing route that changes how older Omni Reference tutorials should be read. The Edit Model can use up to four reference images and can also support targeted inpainting or canvas expansion. For a small retailer, the problem is not just getting an attractive image: the pictured object must remain recognizably the same. This guide uses a fictional blue ceramic bottle, its cap, a tabletop and an outdoor backdrop to explain a controlled process. All examples are conceptual; they are not actual interface captures or successful live renders.
Start by identifying facts the image cannot rewrite
Before uploading anything, record product-defining details that must be right: bottle silhouette, closure shape, blue glaze, label placement and whether liquid is visible. Those details are constraints, not open-ended mood suggestions. Photograph or provide the object from more than one angle where possible and crop the images to emphasize the item without hiding important parts. If the job is a real ecommerce product photograph, generated scenes should be labeled or accompanied by honest real packshots according to the channel’s rules. Do not invent accessories, sizes, materials or product claims just because a rendering looks plausible.
Use the V8.2 Edit Model rather than outdated Omni steps
Midjourney’s official Version page says V8.2 became default on July 24, 2026. Its Edit Model replaces Omni Reference and Character Reference in V8.X; Omni Reference remains a V7 feature and should not be treated as the default current route. On the website, add images in the “Attach to prompt” area, or use the documented –edit URL syntax in Discord. The Edit Model accepts up to four references. More images do not necessarily improve fidelity: every extra reference can introduce a competing angle or incompatible color. Choose each image for a defined role, such as front view, side profile, cap detail or background concept. For the source-specific control described here, consult version compatibility table.

Give each reference a clear job in the prompt
In a multi-reference setup, describe the intended result in plain language. Example: “Place the same blue ceramic bottle from the reference photos on a wooden café table in soft morning light; keep the bottle profile, lid and label unchanged.” A second prompt might ask for a front-facing studio scene. Document which file represents the product and which represents the environment, then review whether the output accidentally borrows background objects into the product. If two product photos disagree because of lighting, describe the approved real color in words. Avoid uploading private client assets without authorization, and do not assume that model guidance amounts to product identity verification.
Control framing without relying on the default aspect ratio
The Edit Model initially tries to match the first uploaded image’s aspect ratio. If the destination is a landscape banner, specify the target shape with –ar; your default aspect ratio setting is not automatically inherited. Compose with enough negative space for the site’s real headline and CTA, and reserve room for potential mobile cropping. Do not put a price or certification badge inside the generated pixels. Create two deliberate crops, a square product card and a wide editorial header, and compare how much of the product is lost. Export and inspect both crops rather than stretching a square generation across a wide slot.

Target a local edit instead of regenerating everything
Use the Editor for a faulty reflection or a distracting prop rather than changing the entire image. Official Edit Model guidance describes targeted inpainting and expansion, with the newer model handling these edits on V8.X. Make one change at a time: remove the extra cup, correct a visible background object, or add clean space around the bottle. Keep the unedited original adjacent to the changed candidate. After the edit, verify that the real product itself did not subtly change shape, closure or lettering. If it did, redo the area using a narrower mask or abandon the generated composite. For the source-specific control described here, consult the current Edit Model guidance.

Build a product-identity review matrix
Evaluate every candidate in four columns: silhouette, distinctive features, label and context. “Pass” means the visual agrees with the source photographs; “review” means a human needs additional evidence; “reject” means a product fact has changed. A pretty result can fail. For the hypothetical bottle, a different cap or altered logo is an immediate reject, while a modest background light shift may be acceptable. Do not give a numeric consistency score unless you actually conducted a defined test with recorded cases. Save the review notes so an approver can understand why a version was selected.
Release a transparent asset package
Deliver the real approved product photograph, any generated lifestyle composite, a statement describing generated elements where needed, the prompt, input source rights and separate final copy. Record whether a model version or workflow was changed during production. The official Edit Model documentation also notes compatibility limitations: Edit Model results are not currently compatible with Remix or –tile. Avoid promising those paths in downstream instructions. For a merchant, the success condition is correct buyer expectations and a consistent usable image library, not simply a large number of attractive versions.
How We Research
We checked Midjourney Edit Model documentation, V8.2 release and feature table and the Editor guide on October 28, 2026. The blue bottle is a fictional sample used to illustrate risk controls, not a real commercial product or tested output. The supporting diagrams are conceptual and do not imply a vendor UI, our original test, an approved endorsement or a measured quality score.
Frequently Asked Questions
Midjourney currently documents up to four reference images attached to the Edit Model.
No. Omni Reference is documented as a V7 feature; V8.X uses the newer Edit Model instead.
Yes. Text, logo placement and object geometry require independent comparison with an approved product photograph.
It initially follows the first reference; use an explicit –ar value to request another aspect ratio.
Not automatically. Keep authoritative real packshots and comply with the marketplace’s image and disclosure rules.
