Midjourney updates V8.2 Edit with 4-image references and --hd output

• Midjourney's first V8.2 Edit test combines instruction-led changes, local image edits, and canvas expansion, according to Midjourney's launch announcement.
• The model accepts as many as four image references at once and is positioned as the successor to Omni Reference, Midjourney's launch announcement says.
• Midjourney says HD mode renders edit runs at 2x resolution, according to a Midjourney support reply.
• A character-sheet recipe from koldo2k turns one reference into a close portrait plus front, side, and back views on a single white canvas.
• Midjourney pushed an image-quality update after the first day of testing, Midjourney's update says.

Midjourney’s release note asks testers to flag failed generations with a sad-face report or submit the job ID or lightbox URL. The company says its alpha.midjourney.com interface will change rapidly while it experiments with an editing-first UI.

Midjourney lists four parts to the V8.2 test:

1. Edit an image with written instructions.
2. Generate with as many as four image references.
3. Brush-select an area for inpainting, or expand the canvas for outpainting.
4. Bring personalization, moodboards, and srefs into the edit run.

The company adds one qualifier: moodboards and srefs may need more prompt direction in this version. Existing Editor documentation describes a workspace that can combine remixing, inpainting, Pan, and Zoom Out.

The four-reference limit turns the editor into a small art-direction stack: a subject, prop, wardrobe treatment, and environment can arrive as separate inputs. Midjourney’s examples combine distinct source images into a single scene.

The Omni Reference replacement has an odd historical wrinkle. DavidSHolz said --oref never ran natively in V8 and instead fell back to V7; Midjourney and Midjourney said the old control remains available by switching to V7.

Character sheets are the clearest early workflow. koldo2k's recipe uses a generated character as the reference, then specifies a seamless white background, one large close-up, and a clean full-body turnaround.

1. Generate the character in the desired style.
2. Attach that image to an Edit prompt.
3. Request front, side, and back views, a character name, and no panels, borders, or redesign.
4. Set the canvas to --ar 16:9.

Midjourney’s release note and DavidSHolz's setup reply identify four ways into the test:

• Drag an image into the prompt bar under “attach to prompt.”
• Open Edit in a lightbox’s lower-right corner.
• Use the left-side Edit tab and upload an image.
• On Discord, type --edit url....

A Midjourney support reply says edit runs can be made at 2x resolution by appending --hd to the prompt or selecting HD from the dropdown.

Midjourney updated the V8.2 Edit model for better image quality and asked people who had trouble during the first 24 hours to rerun their prompts.

The first reactions were uneven. AllarHaltsonen said the V8.2 edit model was “not there yet.”

The official documentation leaves a version boundary unresolved. The Full Editor docs say V8.2 images can enter the Editor, but that its functionality currently uses V6.1.

The Vary Region docs also say HD images edited on midjourney.com return at SD resolution and can then be upscaled, while Midjourney's HD reply says the new edit model can run at 2x. The release note does not specify whether the V8.2 test is a separate path from those documented V6.1 tools.