Midjourney Tutorial: Quick Start with Image-to-Image and Vary Region Partial Redraw

Want to bring a reference image’s composition, character pose, or material/style into your generation workflow? Midjourney’s image-to-image and partial redraw (Vary Region) are very practical. Below, following the order of “use image-to-image to set the direction first, then use redraw to refine details,” the steps are explained clearly so you can avoid detours.

How to start image-to-image: upload an image and use it as a prompt

In Midjourney Web, go to Create. First, drag the reference image into the input box or click upload so it appears in the prompt area. After confirming the image thumbnail has been added, supplement it with text descriptions—such as the subject, scene, camera, and style—so Midjourney knows “what to reference” and “what to generate.”

If you want it to look more like the reference image, you can add the parameter “--iw 1.5” to “--iw 2” at the end of the prompt (the higher the value, the stronger the reference weight). Conversely, if you want to keep the inspiration but not have it look too similar, lower “--iw” and use more specific text to constrain the details.

How to write image-to-image prompts: lock in the key points of composition and style

The most common issue with image-to-image is “it looks similar but it’s not right,” usually because you only gave mood words and didn’t clearly describe the composition and subject relationships. It’s recommended to write in this order: subject (who/what) + action/pose + environmental elements + lighting (soft light / side backlight) + materials (leather / metal / film grain) + visual style (realistic / illustration / photography).

In Midjourney, negative prompts are also important—for example, “no text, no watermark, no extra fingers” can reduce common flaws. If characters become distorted, first delete half of the “fancy adjectives” from the prompt, keep only the most critical structural information, and the results are often more stable.

Vary Region partial redraw: change only one area without starting over

When you select one of the generated results in Midjourney and Upscale it, the interface usually shows a “Vary (Region)” button. Click it, then use the lasso/brush to circle the area you want to modify—such as the face, hands, clothing patterns, or background signs. Try to trace close to the boundary to avoid selecting parts you don’t want to change.

After selecting, add a short prompt in the popup that applies “only to that area,” such as “replace with leather jacket, realistic stitching” or “remove logo, clean fabric.” Midjourney will try to keep the unselected areas unchanged and regenerate only within the region you circled.

Common failure points and how to fix them: more similar, cleaner, more controllable

If you see obvious seams along the edges after redrawing, it’s usually because the selection was too large or the prompt changes were too drastic. Shrink the selection to “cover only what needs to be changed,” and adjust the prompt toward “minor tweaks” rather than “changing the theme.” If details keep drifting off, go back to the image-to-image stage and increase “--iw” to lock in the overall direction first, then do Vary Region for local edits.

One last reminder: Midjourney’s redraw is better for fixing small issues (hands, expressions, local materials, background distractions). Don’t expect to transform an entire image from photorealistic to anime in one go. Following the rhythm of “image-to-image to set the big framework + Vary Region to fix local parts” will be much more efficient.