Midjourney opens V8.2 image-edit model to testing
Midjourney has opened its first V8.2 image-edit model to all users for testing, adding instruction-based editing, multi-image references and inpainting to its main tools.
Midjourney announced that it is opening its first V8.2 image edit model to everyone for testing, and is asking users to help find failure cases before more updates ship.
According to the company, the model supports:
- Editing images from text instructions
- Generating images from up to four reference images at once, which replaces the earlier omni-reference feature
- Inpainting, which changes specific areas of an image, and outpainting, which expands the canvas
- Personalization, moodboards and style references (srefs). Midjourney notes that moodboards and srefs "may work best with some extra prompt direction in this version."
Users can reach the model in several ways: by dragging images into the prompt bar under "attach to prompt," through an "edit" button in the lightbox, through a new "edit" tab on midjourney.com, or with a --edit parameter on Discord. Midjourney has updated the interfaces on both midjourney.com and alpha.midjourney.com. It says the alpha site "will be changing rapidly" as it tries out new interface designs for the model.
The company expects "a lot of edge cases." It asks users to flag poor results with the sad-face emoji in the lightbox or to send job IDs, and to post feedback in its #ideas-and-features Discord channel. It promised "a lot of fast-follow updates."
The announcement includes no benchmarks, no comparisons with other editing tools and no timeline for a final release.