Midjourney September 2026 · Edit Fixes, --tile and Style Previews

Quick Summary

• Midjourney officially improved inpainting and outpainting so targeted edits change only selected pixels and repeated edit passes hold quality more reliably.
• The Edit model can already change perspective through written instructions. Woollyfern reports that easier camera-control tools are in development, based on an office-hours preview, but they are not a released control panel yet.
• Midjourney officially fixed faint seams in new V8.1 and V8.2 images created with --tile. Existing tiles are not regenerated automatically and may still need post-processing.
• The alpha website now offers live previews that apply the latest prompt to liked, featured or searched SREF styles before the user submits a full job.
• The alpha interface also includes edit history inside the image lightbox, a Change Style experiment, default parameters, a wider Create feed and several mobile improvements.
• V9 training, Niji 8, experimental upscalers, easier camera control, a future video model with audio and a full mobile app are roadmap reports. No dependable release date is available in the cited material.

Credit and evidence status

This article is based on Woollyfern's embedded September 2026 update, published on September 30, the supplied transcript and Midjourney's official September announcements and documentation. Woollyfern, also known as Stephanie, demonstrates the interface and translates office-hours discussions into a concise creator-focused update.

The status labels matter. Edit-model improvements, live style previews, alpha interface changes and the V8.1/V8.2 tile repair are official shipped or active alpha features. Easier camera control, V9 progress, Niji 8 features, experimental upscalers, a future video model and a full mobile app are reported previews. They may change before release.

The related Woollyfern videos embedded below are practical companion guides. They cover SREF, Personalization, Style Explorer, Style Creator, V8 and parameter workflows in more depth than a monthly news roundup can.

Edit model: cleaner inpainting and stronger sequential edits

Midjourney launched its V8.2 Edit model in late August. It combines instruction-based editing, generation with up to four reference images, inpainting and outpainting. It replaces the older Omni Reference and Character Reference routes for V8.2 workflows, while also supporting Style References, Moodboards, Personalization and HD output.

The September fix addresses a practical failure mode. Earlier inpainting could alter more than the erased area. Midjourney now says targeted edits should change only selected pixels. The company also says repeated edits should preserve image quality more effectively, which matters when a creator changes wardrobe, pose, prop and environment across several passes.

Keep the mask as small as the requested change allows. Name what should change and what must stay fixed. After each pass, compare the face, hands, clothing, background geometry, lighting and aspect ratio against the previous approved frame. The model is more controlled, but it still needs review.

Midjourney's official documentation also shows perspective changes through instructions such as front and rear views. Woollyfern reports a dedicated, easier camera-control approach shown during office hours. That preview is promising, but it should not be described as a generally available camera tool yet.

The --tile seam fix for V8.1 and V8.2

The --tile parameter makes opposite edges line up so one image can repeat across fabric, wallpaper, packaging or a digital background. Midjourney confirmed that new V8.1 and V8.2 tile generations now blend without the faint border that some users saw between repeats.

Add --tile at the end of the prompt, generate the square tile, then inspect it as a repeated grid rather than trusting the single image. Midjourney's website can preview the repeat when you open a tile image. The documentation warns that upscaling can break the seamless edge, so test again after any resize or enhancement.

The fix applies to newly generated images. Existing affected tiles do not repair themselves. Woollyfern links a separate post-processing tutorial for older work. If the pattern is for print, test the repeat at production resolution and check every boundary before sending it to a manufacturer.

The Edit model itself is not currently compatible with --tile, according to Midjourney's documentation. Create the pattern with the normal image model rather than assuming an edited tile will preserve seamless edges.

Live SREF previews turn style search into a visual comparison

The alpha Create page now has a Styles sidebar with a Live previews switch. Midjourney applies the latest prompt text to small thumbnail previews across liked and featured styles. Scrolling loads more choices, and selecting a thumbnail submits the prompt with that style.

The feature also works when searching for a specific SREF code. This changes style exploration from a code-first task into a visual comparison: keep the content prompt fixed, scan several aesthetics and spend a full generation only on the strongest direction.

Preview images use the prompt text by itself, so treat them as directional. A final result can still change when image references, weights, aspect ratio, stylize, personalization or other parameters are added. Midjourney describes the fast previews as an experiment on alpha.midjourney.com.

A Style Reference controls visual treatment such as color, medium, texture and lighting rather than copying the subject. Personalization learns broader aesthetic preferences from selected images, while Moodboards define a style from a curated collection. Use the tool that matches the level of control you need.

SREF, Moodboards and Personalization are related, not interchangeable

Use SREF when one image or code expresses the look you want. Use a Moodboard when a group of references defines a broader art direction. Use Personalization when you want Midjourney to learn your recurring taste from image selections and apply that preference across prompts.

The live preview feature currently focuses on liked, featured and searched SREF styles. Woollyfern would like similar prompt-aware previews for Moodboards and Personalization profiles, but that extension is a creator recommendation rather than a released feature.

Keep the content prompt simple when testing a Style Reference. Conflicting style adjectives can fight the reference. Compare one variable at a time, save the winning code and record the model version because older SREF codes can behave differently across version changes.

What changed on the alpha website

The alpha site is Midjourney's testing ground, not the safest choice for an everyday production workflow. Its September changes show where the interface may be heading, but individual experiments can move, break or disappear before reaching the main website.

The V8.2 Edit model is built into the image lightbox. Users can open an image, write an edit instruction, attach up to four references and review edits from the same session in one place. The experimental Change Style control offers a faster way to apply a style direction from that view.

Later September updates added saved default parameters, a wider masonry Create feed, restored hover playback for video, project-feed views, in-place organization, persistent search state, clearer edit labels and several performance fixes. Editing canvases can also use HD output.

Mobile and tablet support received an early layout pass, with full-width pages, safer menus, orientation fixes and better upload controls. Midjourney still recommended the production site instead of alpha on mobile, and Woollyfern reports no release date for a full-featured mobile app.

Roadmap watch: camera control, V9, Niji 8, upscalers and video

Camera control: the current Edit model can respond to perspective instructions, but Woollyfern says Midjourney previewed an easier control method during office hours. Treat it as work in progress until an official release explains access and limitations.

V9: Woollyfern reports that training is underway and that the team wants more time to improve it and integrate newer data. Niji 8 is also described as approaching release with unrevealed Niji-specific features. Neither item has a dependable public date in the cited official September announcements.

Upscaling: the report mentions renewed testing, including a 20K proof of concept. A proof of concept is not a promised production option. Midjourney's current V8.1 and V8.2 workflow already supports 2K HD output, which remains the practical choice for most work shown in the video.

Video: Midjourney is thinking about a model after V9 with higher quality and audio while keeping the workflow affordable. The company has not published a September launch date. Cost matters because video jobs consume much more compute than images.

The safest planning rule is simple: use shipped tools for client commitments and treat office-hours previews as research. Save promising ideas, but do not build a deadline around a model or interface that has not launched.

A practical September 2026 workflow

1. Generate or choose a strong base image in V8.2. Decide which details are locked before opening the Edit workflow. 2. Make one targeted instruction at a time and keep the mask narrow. Compare identity, composition and lighting after every pass.

3. For pattern work, return to the normal image model and add --tile. Preview the repeat, avoid upscaling before the seam test and export only after checking a large grid. 4. For style discovery, keep one content prompt fixed and use alpha Live previews to compare liked, featured or searched SREF codes.

5. Move the winning style back into a controlled production prompt with the required aspect ratio, model, quality and reference settings. 6. Record the prompt, model version, SREF code, Personalization code and edit sequence so a result can be reproduced later.

7. Keep alpha experiments and office-hours previews out of contractual deliverables until Midjourney ships them on a stable surface. The monthly roundup is useful for direction, but the official update page and documentation should remain the final reference for current behavior.