This time we’ll focus on Midjourney’s newer, more “controllable” features: using reference images to lock in a style, using character references to keep a person consistent, and combining them with Style Codes to quickly reuse results. These turn Midjourney from “gacha-style image generation” into “plan-driven image generation.” Below, we’ll explain in practical steps how to apply each feature.
Style Reference: Fix Midjourney’s art style
Style Reference is ideal for tasks that need a unified aesthetic, such as brand posters and series illustrations. On the Midjourney web app, drag the reference image into the prompt input box; or in Discord, paste the image link, then add --sref in the prompt to point to the reference image.
If you find Midjourney only “picks up a hint of the vibe,” you can increase the style weight parameter --sw; conversely, if you want more freedom, lower --sw. It’s recommended to choose reference images with a strong style but a simple subject—e.g., works with consistent texture, brushwork, and color palette—so Midjourney more easily learns the style rather than specific content.
Character Reference: Stop Midjourney’s characters from changing faces every image
Character Reference addresses the problem where, across multiple generations, “the same character turns into a stranger.” The method is similar to Style Reference: prepare a clear reference image of the character, add --cref in the prompt, and let Midjourney use that image as the baseline for the character’s appearance.
To make it look more like the original character, increase --cw (character weight). To allow larger variations in pose and clothing, lower --cw. Try to use a single-person image with a clear, unobstructed face and normal lighting; group photos or heavily filtered images make it hard for Midjourney to capture stable features.


