Midjourney Edit · August 2026 Model Guide and News

Quick Summary

• Credit: Video, examples, and first impressions belong to Woollyfern. Document verification, status update and supplemental prompts are from JQ VISIONS.
• Midjourney released the first Edit template for V8.2 on August 27, 2026. The template edits by instruction, combines up to four images, and supports inpainting and outpainting.
• Quick Edit is the simplest input to change an existing image. Place the image in Attach to Prompt, where an instruction such as changing just the color of a shirt is written.
• Open Editor is best suited for localized changes, canvas expansion, and aspect ratio changing. The Edit tab adds layers to position multiple images before generating.
• In Discord, the workflow uses the --edit parameter followed by one or more public URLs. Current documentation allows up to four images separated by spaces.
• The model integrates functions previously associated with Omni Reference and Character Reference. Woollyfern observed better consistency than OREF in V7, but this is an assessment of their testing, not an official success rate.
• Generations are new reconstructions. Pose, face, color, texture or small objects may change even when the request appears localized.
• Current documentation confirms compatibility with V8.1, V8.2, Image Prompts, Style References, Moodboards, Personalization, and --hd. There is no current support for --tile, and Edit results cannot be used with Remix.
• The Alpha interface remains experimental. After the video, Midjourney published new fixes on September 2nd and 16th, including a lightbox editor, mobile improvements and specific messages for editing.
• V9, a possible V8.3, Niji 8 and Video V2 appear in the video as a roadmap reported by Woollyfern. Without an official launch announcement in the sources consulted, they should not be treated as confirmed dates.

Credit, context and state of information

The video is presented by Woollyfern, aka Stephanie, and summarizes the changes to Midjourney during August 2026. Examples of characters, style, clothing and visual correction are tests by the creator. This article incorporates and credits the original video, compares the demonstrations with public documentation, and adds a reproducible workflow.

The editorial date of the article follows the publication of the video on YouTube, September 1, 2026. The update reflects documentation consulted on September 17. This difference is important because the Alpha interface and the Edit model received corrections after recording.

The general capabilities, compatibilities and limits below are official information when linked to Midjourney documentation. Quality preferences and comparisons with OREF are Woollyfern's first impressions. The operational plan is analyzed by JQ VISIONS.

What is the new Midjourney Edit

Midjourney announced the first Edit template for V8.2 on August 27th. The proposal combines three operations: modifying an image with instructions, generating a composition from up to four references and editing specific areas or expanding the canvas through the Editor.

This brings together functions previously dispersed between Editor, Omni Reference, Character Reference and Retexture. Instead of just describing the final image, the user can give a direct order: change the shirt to red, show the character from the back, combine a knight with a horse or convert the image into a pencil drawing.

Edit doesn't literally change the pixels like a traditional application. The system generates four new images guided by references and instructions. This distinction explains why a small change can produce minor changes to the face, pose, light, or color.

The five ways to enter the Edit workflow

Woollyfern organizes the interface into five practical inputs: Quick Edit in lightbox, Open Editor, drag and drop for Attach to Prompt, the Edit tab and the --edit parameter in Discord. Not all serve the same type of work.

Quick Edit and the prompt bar focus on speed. Open Editor favors spatial control. The Edit tab is the layered option for combining and positioning fonts. Discord is useful for those who already operate using commands and have valid URLs for images.

The official documentation presents these entries with slightly grouped names, but confirms the same system: Imagine Bar, Creation Actions, Editor and Discord. Choosing the correct input reduces steps and avoids using a global regeneration for a problem that needed masking.

Quick Edit and Attach to Prompt: the shortest route

Open an image in Create or Organize and choose Quick Edit. The image moves to the Attach to Prompt section in Imagine Bar. Type a direct instruction and submit. In the example, changing the shirt color to red keeps the pose and division very close to the origin.

You can also drag an image from the library directly to Attach to Prompt. The current documentation allows you to fix references with the padlock icon when you want to reuse them in multiple instructions.

To increase precision, separate change and invariants: only change the blue shirt to red; maintains character, face, pose, hands, division, framing and light. This does not mathematically block the remaining elements, but gives the model observable priorities.

Open Editor for inpainting, outpainting and aspect ratio

Open Editor loads the image into the editing environment. It's the best choice when you need to erase an area, regenerate just a region, move or resize content, expand the frame or change the aspect ratio.

In the Editor, the transparent or checkered area communicates where the model should generate. The mask remains a generative orientation, so check edges, shadows, reflections, texture continuity, and objects partially included in the selection.

For outpainting, first position the image on the new canvas and then describe what should be in the additional area. Explicitly preserve lens, perspective, light direction, and depth to prevent the expansion from looking like a different image pasted onto the original.

Edit tab, layers and Discord

The Edit tab provides layers. You can import multiple images, change scale and position, choose the active layer, erase the necessary areas and flatten the composition when submitting an edit. It is particularly useful for mockups, collages and combinations whose geography needs to be decided before generation.

In Discord, add --edit to the prompt and then the referral URLs. Images need to be online. For multiple references, separate URLs with spaces and keep the total within four.

The choice between web and Discord is operational. The web makes it easier to drag, view layers and work with masks. Discord favors reproducible commands and workflows already based on URLs. The underlying model is Edit, not a different capability on each surface.

Four references for characters, clothing and objects

The limit of four images allows you to combine character, clothing, object and environment in a single order. You can also use different views of the same character to reduce missing information when the camera switches from front to back.

Woollyfern considers character consistency to be a marked improvement over OREF in V7. The documentation confirms that Edit incorporates Omni Reference and Character Reference functions, but does not publish a preservation percentage. The correct conclusion is observed improvement, not guaranteed identity.

Assign roles: image 1 defines identity, image 2 defines clothing, image 3 defines object and image 4 defines scenery. When attributes are mixed between two characters, Midjourney herself recommends bringing them together in a single reference image that already shows the desired relationship.

Change style with SREF, Moodboard or attached reference

The Edit model absorbs the Retexture function. You can add a Style Reference, Moodboard or Personalization profile and write an instruction like change the style of this image. You can also attach two images and ask the first to adopt the style of the second.

Woollyfern notes that an attached reference can work better than an SREF in some cases, but may import unwanted content. In her example, a lightning bolt from the style reference appears in the result. That illustrates the difference between learning a visual style and copying a motif.

The documentation recommends describing the style in text to enable aesthetics with greater consistency. Indicate medium, palette, texture, contrast and light. Use --raw when you need to reduce additional creativity and give more weight to the literal statement.

Quality, drift and edge cases

In Woollyfern's tests, the overall quality is high and the original structure survives better than in the old editor. Dotted textures, too dark images, excessively smooth skin, incoherence and differences between the same SREF in a normal generation and in an edition still appear.

Midjourney updated the model's quality on August 29 and asked for failed cases to be repeated. The company also requests feedback via the reaction icon in the lightbox or through the job ID and URL. This confirms that the release is still in a rapid fix cycle.

Save the origin, instruction, references and each approved result. Compare face, hands, object count, text, palette, framing, and product details. If a simple edit fails, downshift, use a mask, or revert to the previous version rather than accumulating drift.

Compatibility and documented limits

Current documentation indicates compatibility with versions 8.1 and 8.2, Image Prompts, Style References, Moodboards, Personalization and --hd. The format of the first reference is used as the default aspect ratio, but can be replaced with --ar.

The template is not currently compatible with --tile. Images produced by Edit are also not currently compatible with Remix. These limitations are later or more specific than the demonstration and should be double-checked before building a process dependent on them.

Compatible does not mean equivalent. A SREF can have different intensity in an edit, HD can reveal artefacts that were not visible in the preview and four references increase the possibility of mixing attributes. Test your real case before promising consistency to a client.

The Alpha website has already changed since the video

In the August summary, Woollyfern shows a redesign with prompt bar, definition pills, sidebars and menus. The recommendation is to use alpha.midjourney.com as a test space, not as a main production environment.

Midjourney confirmed this experimental nature in the August 20th changelog. On September 2nd he added Edit to the Alpha lightbox and a history of session edits. On September 16th, more corrections were published in the editor, specific messages when attaching images and mobile and tablet improvements.

Those who watch the video can now find different labels, positions and behaviors. Learn the mental model, not the exact position of each button: attach references, declare the change, choose global or localized editing, generate and review.

V9, possible V8.3, Niji 8 and Video V2

According to the update reported by Woollyfern, V9 training was temporarily paused to improve the dataset and the effort focused on Edit also delayed the timeline. The creator mentions the possibility of an intermediate V8.3, probably related to data within the V8 infrastructure.

The video describes Niji 8 as next and places a Video V2 after V9. In the official public sources consulted for this article there is no launch announcement with a date for these three models. Therefore, these points remain roadmap reported, not availability confirmed.

For production decisions, use what is available and documented today. Do not delay a project for a speculative date and do not build a commercial offering on capabilities that have not yet been launched. Reevaluate the roadmap when Midjourney publishes an official announcement.

Practical workflow: from approved image to final result

Start by duplicating or saving the approved image. Define a single change and write the invariants. Choose Quick Edit for a simple global change or Open Editor for a region. Attach only the necessary references and describe the role of each one.

Generate four options and reject structural flaws first. Then compare identity, geometry, color, text and light. Continue only from the best version. For independent changes, create branches from the source rather than applying a long chain of edits.

Finalize in HD only after approving composition and content. Save prompt, parameters, SREF codes, references, job ID and model version. This recording transforms a visual experience into a process that can be audited, repeated and delivered.