Beginner tutorials

01 · Enter a prompt and generate images

Midjourney starts with a prompt: a short description of the image you want. On midjourney.com, open the Create page, type your description in the Imagine bar, and press Enter. A standard job returns four different interpretations of the same idea.

In Discord, enter the /imagine command, add your text in the prompt field, and send it. Clear visual details usually matter more than long instructions. Describe the subject, environment, lighting, mood, and composition you actually want to see.

• Website — use the Imagine bar on the Create page
• Discord — use /imagine and enter the description in the prompt field
• Prompt structure — subject, setting, composition, lighting, style, and important details

TipEnglish descriptive phrases are a dependable choice, but clarity is more important than length. Start concise and add only details that affect the image.

02 · Choose a candidate and refine it

Compare the four candidates and select the image closest to your goal. On the website, open the image to access Vary, Remix, Edit, and related controls. In Discord, the V buttons generate four new variations that keep the selected image's general composition and style.

In the latest Discord workflow, U1–U4 primarily separate the selected image from the grid and reveal more tools; they are not the same as the old one-click upscale buttons. Use Vary (Subtle) for small changes, Vary (Strong) for a wider reinterpretation, or rerun the original prompt when the overall direction is wrong.

• Keep the subject and composition when the direction is already right
• Change one variable at a time so you can judge what improved the result
• Use a new prompt or reroll when none of the four candidates fits the goal

TipMidjourney becomes easier when you treat each generation as a controlled comparison instead of trying to make the first result perfect.

03 · Edit, increase resolution, and export

Use the web Editor to change selected regions, remove unwanted objects, replace a background, or extend the canvas with Pan and Zoom Out. Midjourney's current Editor can combine Remix, inpainting, panning, and outpainting. Check the edited image again before final export because regenerated areas can introduce new details.

With V8.1, a square SD image is typically 1024 × 1024 pixels, while HD produces a native 2048 × 2048 image. Midjourney's current documentation estimates about 0.8 GPU minutes for a four-image SD job and about 1.3 minutes for HD. Editing or extending an HD image can return it to SD, so upscale again after the edit when you need a final 2K file.

• Explore quickly in SD while deciding on the subject and composition
• Finish edits before spending extra GPU time on the final resolution
• Use HD or an upscale option for posters, print, and other high-resolution output
• Inspect faces, hands, text, repeated patterns, and image edges before download

TipThe efficient workflow is two-stage: explore several SD options first, then switch to HD only for the final candidate. GPU costs and interface labels may change as Midjourney updates.

01 · The tea-party gentleman

The rabbit head and tailored human body continue a long satirical tradition of anthropomorphic portraiture. A photographic subject placed against gold, neon pink, and petrol blue creates a deliberate photomontage seam: respectable Victorian posture collides with Dada-like collage energy.

TipBind the photographic treatment to the subject, then describe the background as a separate collage layer. This keeps the head from dissolving into the graphic field.

02 · Rialto Bridge

Flat color, bold outlines, a balanced vertical layout, and visible halftone texture make the scene feel like a twentieth-century travel poster. Halftone is especially important: it turns a clean digital gradient into something that reads as ink on paper.

TipFor a convincing vintage poster, combine one landmark, one transport motif, flat colors, coarse outlines, and a clearly named print texture.

03 · Ice-cream disaster

The image holds attention through controlled contradiction. Mint green, strawberry pink, and soft clouds promise sweetness, while eyeballs, slime, and fangs introduce danger. Because cute color remains dominant, the horror reads as a sharp accent rather than the whole genre.

TipKeep the pastel world as the base and use teeth, eyes, or slime sparingly. Reversing that balance produces conventional horror instead of pop surrealism.

04 · The screentone cat

The girl and wallpaper use clean cel-shaded color, but the cat alone receives visible manga screentone dots. Two printing logics coexist in one frame, creating depth that feels assembled rather than uniformly filtered.

TipAttach a texture to a named object—such as ‘the cat rendered with visible halftone screentone dots’—instead of placing the texture in the global style list.

05 · The clay rocker

Caricature works by amplifying the features that separate a face from an average one. Oversized glasses, a long nose, and a narrow face make the reference instantly legible, while matte clay and vinyl-toy cues soften the distortion and prevent it from feeling grotesque.

TipChoose two or three signature features to exaggerate, then add a friendly material such as clay or vinyl to control the emotional tone.

01 · Define what the image needs to do

Do not begin by stacking style keywords. First define the image's purpose, subject, and intended mood. These choices determine the composition and details you need later.

For example, the same coffee cup needs clear materials and negative space in an ecommerce image, while a brand poster depends more on atmosphere and a memorable visual idea.

• Purpose: avatar, cover, poster, or product image
• Subject: the most important person or object in the frame
• Mood: calm, energetic, mysterious, or light

02 · Use a six-part prompt

Describe the subject, environment, composition, lighting, visual style, and key details—in that order. Give each part one job so the model can identify what matters.

TipKeep the first version concise. Too much information can pull attention away from the subject.

03 · Check only three things in the first result

Do not get stuck on small flaws yet. Use these three questions to decide whether the direction is right. If it is not, revise the prompt; if it is, continue refining.

• Is the subject immediately recognizable and positioned for the intended use?
• Do the light and color communicate the target mood?
• Does the image have enough depth and negative space?

04 · Change one variable at a time

If you change the subject, camera, color, and style at once, you cannot tell what made the image better or worse. Keep what works and replace only one variable in each round.

A useful order is subject and composition first, lighting and color second, then materials, clothing, and environmental details.

• Subject feels weak: remove secondary objects and clarify its position or action
• Image feels flat: specify shot size, camera angle, and foreground–background layers
• Mood feels wrong: define the main color, light direction, and whether the light is soft or hard

05 · Run a final check before export

Choose the result closest to your goal and upscale it. Before commercial use, manually inspect faces, hands, text, brand marks, and the structure of small objects.

• Does the aspect ratio and resolution fit the publishing channel?
• Do subject edges, facial features, and repeated textures look natural?
• Did you save the final prompt and version so the result can be reproduced?

01 · Translate feelings into visible details

Words such as “premium,” “dreamy,” and “textured” are difficult to execute directly. Break the feeling into color, material, light, and spatial relationships for more consistent results.

TipThis gives the model more visible information than “a premium, dreamy vase.”

02 · Give the subject an identity, state, and action

Writing only “a girl” produces an extremely broad range of results. Add an age cue, clothing, action, and gaze, but avoid assigning too many competing identities at once.

• Identity: botanist, dancer, or architect
• State: just stepped out of the rain or sorting specimens
• Action: looking back, bending down, or holding an object in both hands

03 · Use shot size and camera angle to control the story

Shot size controls how close the viewer feels to the subject; camera angle changes the subject's psychological presence. Choose one main camera treatment.

• Close-up: emphasizes expression, material, and detail
• Medium shot: balances action and environment
• Wide shot: emphasizes the scale of the space
• Eye level feels natural, a low angle feels powerful, and a high angle feels observational

04 · Define the light's direction, softness, and color

Instead of writing only “cinematic lighting,” explain where the light comes from, whether the shadows are soft or sharp, and which colors appear in the environment.

05 · Put important information first and remove conflicts

Lead with the subject and core scene. Check whether the prompt combines incompatible directions, such as photorealism with flat illustration or hard midday sun with soft overcast light.

• Keep one main direction for each type of information
• Remove trendy words that do not support the image goal
• Separate modules with commas so they are easy to replace

06 · Build your own controlled comparisons

Keep the subject and scene fixed, then replace only the camera, lighting, or material. Save each result with its prompt. After several rounds, you will have a vocabulary that is more useful than a generic keyword list.

• Group A compares only shot sizes
• Group B compares only dominant colors
• Group C compares only lighting directions
• Tag reusable phrases from successful results