Midjourney Tutorial: Detailed Steps for Uploading a Reference Image on the Web and Reverse-Engineering the Prompt

If you want Midjourney to get closer to the image in your head, the fastest way is to “use an image to guide an image.” This Midjourney tutorial focuses only on the web version: how to upload a reference image, use Describe to reverse-engineer prompts, and then keep generating by using the reference image as an image prompt—while also clearing up common stumbling blocks along the way.

1. Enter the Midjourney Web App and Prepare Your Materials

After opening the Midjourney official website and logging in, go to the “Create” page to start generating and managing your works. It’s recommended to prepare a clear reference image first: keep the subject as centered as possible, avoid lighting that’s too dark, and steer clear of heavy watermarks and excessive compression.

If you’re trying to learn a “style,” choose an image with rich visual elements but a clearly defined subject; if you’re trying to learn “composition,” pick an image with an obvious viewpoint (top-down, low-angle, wide-angle), which makes it easier to reverse-engineer usable prompts.

2. Reverse-Engineer Prompts with Describe: From Image to Editable Text

In the Midjourney web app, find the “Describe” entry point and upload your reference image. The system will provide multiple clickable prompt candidates. Don’t rush to use them as-is—prioritize the one that includes “subject + setting + lighting + lens/art style.” The more complete the information, the more stable the results will be later.

After copying the candidate prompt into the input box, it’s recommended to make two small tweaks: first, add the subject details you actually want (clothing material, age range, expression); second, remove elements you don’t need (such as extra background objects or art movements you don’t want). This is the part of a Midjourney tutorial that most clearly separates good results from mediocre ones.

3. Use the Reference Image as an Image Prompt: Make the Output “Move Toward the Reference”

Go back to “Create,” add the reference image in the input box as an image prompt (after uploading on the web, the image will automatically become part of the prompt). Then add your own text prompt: use one sentence to state “what I want to draw,” and then use short phrases to supplement “environment, lighting, style, details.”

If the result looks too much like the reference and lacks variation, reduce your reliance on it: switch to more abstract descriptions and use fewer detail words; if the result drifts too far away, do the opposite—reinforce the subject and key characteristics (e.g., hairstyle, color palette, lens focal length) and keep the prompt more “specific.” When using Midjourney to follow a reference image, a “get it similar first, then refine” rhythm usually saves the most time.

4. Common Issues: Upload Failures, Inaccurate Reverse Prompts, and Off-Track Outputs

The most common causes of upload failures are files that are too large or incompatible formats: convert the image to JPG/PNG first, and compress it to a smaller size before trying again. If Describe generates a very “empty” prompt, it’s usually because the subject in the reference image isn’t clear enough—switching to a clearer image with less obstruction will improve things immediately.

If the output goes off-track, first check whether your prompt contains terms that “fight” each other (for example, wanting both “photorealistic” and “cartoon”). In Midjourney, reducing conflicting terms and placing the most important requirements in the first half of the prompt is often more effective than repeatedly rerolling.