Midjourney's Inpainting is SUPER Impressive!

Midjourney’s long-awaited inpainting feature is rolling out inside Discord, and early tests suggest it can edit selected regions while preserving the rest of an image’s style, lighting, and character consistency. The workflow hinges on enabling “Remix mode” via /settings, generating an image as usual, then using a new “Vary (Region)” option on upscaled results to open an inpainting editor with selection tools (rectangle and lasso) plus an undo button. With that setup, users can prompt for localized changes—like adding a face to a lemon, swapping clothing, or rebuilding missing body parts—without losing the original scene.

In simple demonstrations, Midjourney handles small, targeted edits reliably. A lemon on a sandy beach can be given a realistic, smiling face while the background remains consistent. The same image can then be iterated again: a Lincoln-style top hat is added, and the system continues to maintain the underlying image quality rather than degrading as more edits accumulate. More ambitious prompts—like generating anthropomorphic legs and feet—show a mixed hit rate, with some outputs producing anatomically odd or “corn-like” feet. Yet the feature improves when prompts focus on specific, human-like details; comically large buff arms and hands often come out better than expected, including cases with five fingers and convincing transitions that preserve the lemon’s texture.

The most telling stress test involves a complex, character-specific edit: a photo of Walter White in a yellow hazmat suit eating a burger in a McDonald’s setting. Midjourney struggles with consistency in some iterations—hands may not match the intended burger-holding pose, and the environment can drift toward “subway-like” visuals. Still, the inpainting tool can remove unwanted elements by clearing the prompt and pressing enter, and it can rework hands and objects through region-specific prompts (for example, replacing a Big Mac with a cheeseburger). Background replacement also works: the tiled floor and lighting can be adjusted to look more McDonald’s-like, and clutter can be erased with no prompt. The final result in the demo lands close to the intended composition, including a coherent McDonald’s logo and a background that matches the lighting conditions.

Beyond one-off experiments, community examples point to the same core advantage: subtle, localized changes that keep the character and scene intact. Users report being able to swap sunglasses, clothing, and even vehicles or add new context elements while maintaining the original art style and lighting. Compared with alternatives like Adobe’s generative fill, the consensus in these early comparisons is that Midjourney’s inpainting is among the strongest options for image editing that stays “in-family” with the generated output.

Even so, limitations remain. Editing is currently Discord-based, which can be a barrier for non-technical users, and the feature lacks some UI conveniences such as a website interface and more granular brush controls (e.g., circle/magic-wand selection, feathering, and blending controls). Community feedback also highlights occasional unpredictability and the need for better prompt guidance. Still, the overall takeaway is clear: Midjourney’s inpainting is a meaningful leap for existing users, enabling precise fixes and creative refinements without exporting images to other tools.