Superpower Daily
Judge Vacates Pentagon’s Anthropic Blacklist and Orders Defense...Databricks Files $5B Equity Offering With Nearly All of It Alre...Snowflake Open-Sources Semi-Persistence With Sub-Second Single-...Cisco Rolls Out MyAgent to 90,000 Employees for Supervised Cros...Midjourney Opens V8.2 Image-Editing Tests With Up to Four Refer...Gemini Omni 1.1 Flash Adds 40-Second Scene Extensions and 4K Fi...Judge Vacates Pentagon’s Anthropic Blacklist and Orders Defense...Databricks Files $5B Equity Offering With Nearly All of It Alre...Snowflake Open-Sources Semi-Persistence With Sub-Second Single-...Cisco Rolls Out MyAgent to 90,000 Employees for Supervised Cros...Midjourney Opens V8.2 Image-Editing Tests With Up to Four Refer...Gemini Omni 1.1 Flash Adds 40-Second Scene Extensions and 4K Fi...
Modelspublished

Midjourney Opens V8.2 Image-Editing Tests With Up to Four References

The test brings instruction-led edits, canvas expansion, and multi-image generation into one workflow, while Midjourney asks users to identify failures and help reshape the interface.

By 2 min read
Midjourney Opens V8.2 Image-Editing Tests With Up to Four References
Sign in to save and react

Listen to this story

The audio brief

About 1:17
0:001:17
Read transcript
Midjourney has opened broader tests of V8.2, an image-editing model that brings written instructions, selective changes, canvas expansion, and up to four reference images into one workflow. Instead of treating editing as a single isolated feature, the test lets users start with an uploaded image and then reshape it in several ways. Inpainting changes a selected part of the picture, while outpainting extends the canvas beyond its original edges. Midjourney says the new workflow replaces omni-reference, its earlier reference-image feature, with support for as many as four images at once. Personalization, moodboards, and srefs are still available, although the company warns they may need more specific prompting in this version. Access is spread across the product: users can attach images in the prompt bar, choose Edit from an image lightbox, upload through the Edit tab, or use Discord with the command “dash dash edit,” followed by a URL. The model is running on Midjourney’s main website, its alpha site, and Discord, but neither the behavior nor the interface is final. This is explicitly a community test. Users can report failures through the lightbox’s sad-face control, or share the affected job ID or URL. The practical question now is how quickly Midjourney’s fast-follow updates refine both the model’s edge cases and the way people work with it.

Story brief

3 key points

Midjourney’s V8.2 test turns image editing into a broader workflow rather than a single new feature: users can start from an uploaded image or lightbox and combine instructions, selective edits, canvas expansion, and references. The model supports up to four reference images and replaces omni-reference in this workflow, while personalization, moodboards, and srefs remain available with possible extra prompting. It...

  1. 01

    Users can launch edits from the prompt bar, image lightbox, Edit tab, or Discord with `--edit url...`.

  2. 02

    Inpainting changes selected regions; outpainting expands the canvas, bringing both controls into the same V8.2 test.

  3. 03

    Midjourney warns moodboards and srefs may require additional prompt direction in this version.

Midjourney has begun broader testing of its first V8.2 image-editing model, giving users one place to alter an image with written instructions, use as many as four image references, and change or extend portions of a canvas. The release shifts several creative controls into an edit-focused workflow, but Midjourney is also asking users to surface cases where the model does not behave as expected.

One editing model, several inputs

The V8.2 model accepts natural-language editing instructions. It also generates images from other images, allowing up to four references at once. Midjourney says this replaces its omni-reference feature in the new editing workflow.

The model includes inpainting, which changes selected areas of an image, and outpainting, which expands its canvas. It also supports personalization, moodboards and srefs. Midjourney cautions that moodboards and srefs may need extra prompt direction in this version.

Ways to access the model

Users can reach the model by attaching images to the prompt bar, selecting Edit from an image lightbox, or using the Edit tab to upload an image. It is available on Midjourney’s main website and alpha site, as well as through Discord.

Discord syntax
--edit url...

The multiple access routes place editing alongside Midjourney’s existing prompt and image-viewing surfaces rather than confining it to one new destination. The company has updated the interfaces on both sites, and says the alpha version will change rapidly as it experiments with how the model should be used.

Testing is part of the release

Midjourney is presenting the rollout as a community test, not a settled interface or behavior set. It expects edge cases and directs users to flag poor results through a sad-face control in the lightbox or by sharing the affected image’s job ID or URL while it is open.

The company is also requesting feedback on the user interface and directing ideas and edge cases to its ideas-and-features channel. It says fast-follow updates are planned after the initial test begins, leaving the pace and substance of those changes as the next practical question for users trying the model.

Sources

  1. updates.midjourney.comEdit Model for V8 - Midjourney

Related coverage

Continue reading