Midjourney finally rolls out "inpainting" - customizing AI images after generation
Key Points
- Midjourney is testing a new feature called "Vary Region" that allows you to edit specific areas in images with text commands, for example to change facial expressions.
- The feature works best for relatively large image areas and small adjustments, while basic and fancy changes are more difficult for the system to handle.
- Inpainting is not a "magic fix," Midjourney says, and for larger changes to the image, a full regenerate may be the better way to go.
Midjourney is currently testing a long-awaited feature: You can now edit individual regions of an image via a prompt, a feature typically known as "inpainting".
Midjourney calls its variant of "inpainting", i.e., painting over parts of an image, "Vary Region". This option appears on upscaled images and is next to the existing Remix functions "Vary Strong" and "Vary Subtle". You need to make sure that Remix Mode is enabled in the settings (prompt "/settings"). Otherwise, the button will not show up.
If you press "Vary Region", a new interface will appear. Here, you can select an area in the upscaled image and edit it with text prompts. For example, you could use a prompt to make a portrait photo with a sad expression smile.
Of course, you can go on by upscaling the newly created image and further customizing the prompt. For example, you can add a hat to the portrait image.
Inpainting isn't a silver bullet for image editing
According to Midjourney, inpainting works best with relatively large areas of the image, 20 to 50 percent of the image, and when the prompt is only slightly adjusted. Replacing a small detail in the background is likely to be more difficult.
Also, inpainting works more reliably when the prompt change blends well with and complements the original image, as in the example above. The more basic and outlandish the image change, the more difficult it is for the system to implement.
According to Midjourney, inpainting is not a "magic fix" and often the "vary subtle" feature would work better. This feature generates new images based on the original image and a customized prompt, rather than just changing an area. In the cat example, however, this didn't help.
For major changes in the image, such as the cat on the red-bearded man's head, re-creation may be the way to go. In this case, Midjourney will create a new composition from scratch that better fits the prompt. Of course, you will lose the look of the original creation.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now