Midjourney starts V8.2 image-editing tests
Midjourney says V8.2 adds instruction-based edits, image inputs, brush inpainting, and outpainting. The tools retain personalization, moodboards, and style references.
TL;DR
- The V8.2 test joins instruction-led edits, generation with other images, brush inpainting, and outpainting, according to the V8.2 announcement.
- Personalization, Moodboards, and style references carry into the edit pass, the announcement says.
- Early creator tests are already prompting over an original output and changing style between iterations, as lloydcreates's test puts it.
- Creator workflows are splitting image jobs: 0xInk_ assigns concept art to Midjourney and character-sheet cleanup to GPT 2 in a reply.
Midjourney’s Version page says V8.2 became the default model on July 24, with a focus on aesthetics, quality, and Personalization. Its Style Reference docs define the control as carrying colors, medium, textures, and lighting from an image, rather than its people or objects.
The V8.2 edit test
The four operations in the V8.2 announcement are:
- Editing with instructions
- Generating images with other images
- Brush-based inpainting
- Outpainting
Three ways into Edit
DavidSHolz gave three entry points in a reply:
- Drag images into the prompt bar.
- Choose Edit in the left navigation.
- Choose Edit at the lower right of the lightbox.
Style systems
The test promises to retain Personalization, Moodboards, and style references, the announcement says. Moodboards documentation describes them as curated images used to create a project-specific aesthetic, while Style References are a narrower visual-style control.
carolletta's V8.2 reference-style prompt pairs an --sref code with high style weight and strong stylization:
artistic portrait of a beautiful character --chaos 45 --ar 9:16 --sref 1744656026 --raw --sw 1000 --stylize 600 --weird 25
Iterations from an original output
lloydcreates described prompting and changing styles through multiple iterations that began with an original output.
DavidSHolz said in a reply that maximum Personalization turned front-page images into “crazy singularity space scenes.”
A six-model generation test
ai_artworkgen's six-model comparison is explicitly a pure-generation test, so it does not assess the new editor. Its lineup was:
- Midjourney V8.2
- Grok Imagine 2.0
- Seedream 5 Pro
- Nano Banana Pro
- Reve 2.0
- GPT 2
The creator's working stack starts in Midjourney, then moves to Nano Banana 2/Pro or GPT 2 to build out worlds, because they find Midjourney less consistent, the conclusion says.
V8.2 to video
0xInk_ used V8.2 for an X-Men-inspired Spider-Man character design, then made the swinging sequence in Seedance 2.5.
juliewdesign_ listed Midjourney 8.2, Seedance 2.5, and Topaz Starlight 2.6 in a chase clip; a follow-up set the delivered resolution at 720p in a reply.
The V6.1 editor caveat
Midjourney’s Editor documentation says V8.2 images can enter the Editor, while the Editor's functionality currently uses V6.1. The same page says the existing interface can combine Remix, Vary Region inpainting, Pan, and Zoom Out.
The docs direct external-image uploads and advanced tools to the website’s Edit tab, while a lighter Editor opens from Create and Organize pages. That leaves a concrete implementation distinction beside the V8.2 edit-model announcement: V8.2 is the image version, and V6.1 is the documented editing functionality.