How to Use Midjourney Vary (Region) to Fix Specific Parts of an Image
You generate an image that is almost exactly right. The composition is strong, the lighting is exactly what you wanted, the color palette works. But there is one area that does not fit: a figure in the corner that looks wrong, a foreground element that is too busy, a sky that feels flat.
The old solution was to regenerate and hope you got closer. The better solution is Vary (Region).
What Vary (Region) Does
Vary (Region) is Midjourney's inpainting tool. It lets you select a specific area of a generated image, describe what you want in that area, and regenerate only that section while leaving everything else intact.
The surrounding image acts as context. Midjourney generates the new content to blend with the existing lighting, color palette, and compositional style. When it works well, the result looks like the original image was always that way.
How to Access Vary (Region)
After generating an image and upscaling it, click the Vary (Region) button below the upscaled image in Discord. This opens an editor interface with your upscaled image as the base, a selection tool (rectangular and freehand lasso), and a prompt field at the bottom.
Select the area you want to change, write a prompt describing what you want in that area, and submit. Midjourney generates four variations of the selected region, rendered to blend with the surrounding image.
Selecting the Right Region
Region selection has more impact on the result than most people expect.
Select too small: The model has very little room to work with. The generation often looks cramped or ignores the prompt in favor of filling the space with something generic.
Select too large: The model treats the selected area as a significant compositional element and may generate something that does not blend smoothly with the edges.
Select with tight edges around a specific element: The best approach for most fixes. Select slightly beyond the element you want to change to give the model room to blend, but keep the selection focused on the problem area.
Leave complex border elements outside the selection: Blend quality is better where the selection edge lands in a visually simpler zone.
Writing the Inpainting Prompt
The prompt for Vary (Region) behaves differently from a standard Midjourney prompt. You are describing only the selected area, not the whole image.
Be specific about what should replace the current content: "Dramatic storm clouds with volumetric lighting, deep purple and charcoal gray" rather than "better sky."
Match the visual language of the rest of the image: If your image is a cinematic neon cityscape, write inpainting prompts with consistent language. Do not introduce elements that break the visual register.
Do not repeat the full original prompt: The model already knows the context. You only need to describe the target region. A short, specific inpainting prompt usually outperforms a long one.
