2) Once you have a generation that you like, upscale it.
Reminder: You want a three panel image. Preferably with a large panel and two small panels.
3) With your upscaled image, click on the vary region button. Using the square selection tool, select only one panel. Make sure your selection slightly extends past the edges of the panel.
Important: Only select one, or at most, two panels. Do not select the entire image.
4) Edit only the subject of the prompt and change the stylization to low.
For example, if your subject is a brunette woman, you can change the subject to brunette woman at dinner. Do not change any of the remainder of the prompt.
Important: Add --s 50 to the end.
5) This is where the magic happens. It appears that by selecting one panel for inpainting, Midjourney will reference the subject in the other panels.
If you are having issues, try selecting a different panel and re-running the inpainting prompt. Also make sure to use --s 50.
6) To go a step further, upscale one of the inpainted versions and click the vary region button again. This time select every panel besides the one that you like.
This is less stable, but with some re-rolls you can create multiple images of your subject at the location.
7) If you would like to extend the height and width of your panels, I recommend using adobe's Gen Fill for this task.
This last step is optional.
Thank you for reading along! If you enjoyed these posts,
1) Follow me @Mr_AllenT for more AI content and tutorials.
2) Bookmark, like, and repost to share with your friends!
For this example, I generated a fashion model with Flux on Affogato to use for the outfit swaps. However, you could use a real photograph as well.
Prompt: "A wide full body photograph of a beautiful fashion model that is asian with short bob cut orange hair. The background is solid white. Wearing business dress and heels"