How to Use the Midjourney V8.2 Edit Model

01

What the V8.2 edit test actually is

Midjourney began letting everyone test its first V8.2 image-edit model on August 27, 2026. Midjourney explicitly calls it a test and says the model and interfaces will receive fast-follow updates, so exact labels and paths can change.

The new model is distinct from the older Editor workflow. Midjourney's current Editor documentation still notes that some Editor functionality uses V6.1 even when the source image was created with V8.2, so check the current UI before assuming every Edit screen is already using the V8.2-native model.

02

Four ways to start an edit

• Drag or attach an image in the prompt bar.
• Open an image in the lightbox and choose Edit.
• Open the Edit tab and upload an image.
• In Discord, use <code>--edit</code> with an image URL.

Use the web path when you want visual image selection and editing controls. Discord is useful when your workflow already depends on prompt commands and hosted image URLs.

03

Describe the change instead of rewriting the whole image

When the source already has the right subject and composition, focus the prompt on what should change. A useful structure is: what to preserve + requested change + placement or style constraint.

04

Use up to four image references

Midjourney says the V8.2 edit model can generate with up to four image references at once. That can support combinations such as a base portrait plus wardrobe reference, a product plus environment reference, or a subject plus separate style guidance.

Start with the minimum number of references needed. Each additional image gives the model more visual information to reconcile, which can also introduce conflicts or drift.

05

Use inpainting for local changes

Inpainting is the better fit when only one region should change. Midjourney's Editor exposes selection and erase tools for localized regeneration, while the V8.2 announcement explicitly includes changing specific areas as a supported capability.

After generation, inspect the edges around the edited area. Generative editing can change nearby lighting, texture, geometry or identity even when the prompt targets a small region.

06

Use outpainting to change the canvas or aspect ratio

Outpainting expands the canvas and asks the model to generate content outside the original boundaries. That is useful when a source image needs to become a wider website hero, a horizontal ad or a vertical social asset.

Give the prompt guidance about what should occupy the new space. Otherwise the model must infer the extension entirely from the source composition.

07

Personalization, moodboards and style references

Midjourney says the test supports personalization, moodboards and style references. Its launch note cautions that moodboards and style references may work better with extra prompt direction in this version.

For repeatable campaign work, describe the intended visual treatment as well as attaching the style control instead of assuming a style reference alone will preserve every detail.

08

Review preservation-sensitive details

• Faces and identity
• Hands and fingers
• Logos and product labels
• Text and typography
• Jewelry and small accessories
• Repeated patterns
• Packaging and product geometry

Generative editing is not deterministic layer editing. If a logo, regulated label or exact product detail must remain unchanged, preserve the original asset and composite exact elements back in with a conventional editor when necessary.

09

A practical creator workflow

Start from the approved base image, request one meaningful visual change, add another reference only when needed, repair local mistakes with inpainting and use outpainting last when the final deliverable needs a different frame.