Midjourney's V8.2 edit model lets you change images with plain instructions

Midjourney spent September making its V8 generation of models much easier to steer after the first image comes out.

Edit with a sentence

The V8.2 edit model, which first appeared on 27 August, lets you change an image by describing the change. It handles inpainting and outpainting, and you can attach up to four reference images to guide the edit. Since 2 September it has been available directly in the lightbox, so you no longer need a separate editing step.

Style previews and seamless tiles

• The Styles sidebar can now show what your latest prompt would look like across different styles before you commit.
• You can save default parameters so every new prompt starts the way you like.
• Tile generation for V8.1 and V8.2 now blends seams better, which matters for patterns and textures.

What it means for your workflow

Instruction-based editing is where AI image tools become practical for client work. Instead of re-rolling until a hand looks right, you fix the hand. Reference images help keep a product or character consistent across a set. You'll still need to finish the work in your own tools, but the gap between a first draft and something usable is getting smaller.

As always with generated imagery, check the terms of your plan and your client's policy on AI-made assets before anything ships.