Midjourney opens public testing for V8.2 image editing model
Midjourney
An instruction-driven image-editing model that ships alongside V8.2 to midjourney.com, alpha.midjourney.com, and the Discord client. It edits images by natural-language prompt, accepts up to four reference images to replace the previous omni-reference system, and adds localized inpainting and canvas outpainting. Personalization, mood boards, and style references (sref) compose with the new editor; UI is expected to iterate as the model is rolled out in early testing.
Why it matters
Replaces Midjourney’s pseudo-reference workflow with an actual instruction-editing model and gives the platform its first built-in inpainting/outpainting on top of the V8.2 generator, putting it on a more direct footing with Recraft V4, Adobe Firefly, and Flux-driven editors.
Importance: 2/5
official confirmation