The Complete Midjourney v7 Prompt Guide: Parameters, Styles, and Workflows (2026)

What's New in Midjourney v7

Midjourney v7 is the biggest leap since v5. Released in early 2026, it delivers significantly better photorealism, improved hand and face anatomy, stronger prompt adherence, and a brand-new personalization system that learns your aesthetic preferences after a short ranking session. The default model now produces images with sharper micro-detail and more accurate text rendering, closing much of the gap that previously favored Flux 1.1 Pro and Ideogram for typography. Character consistency has also taken a major step forward: the new --cref engine preserves facial features, hairstyles, and clothing across generations far more reliably than v6. Beyond stills, v7 introduces native video generation through the /video command, which animates any image into a 4-second clip with optional motion prompts. Two other quality-of-life upgrades matter for daily use: Draft Mode, which generates rough previews ten times faster at half the cost so you can iterate on composition before committing to a full render, and an updated Niji 6 mode for anime and illustration styles. If you came from v6, the syntax you already know still works, but the defaults are different — most v6 prompts will look better in v7 once you reduce --stylize values and let the model breathe.

Prompt Anatomy: The Six-Part Formula

Every strong Midjourney prompt follows roughly the same structure: subject, medium, style, lighting, composition, and mood. Think of it as describing a photograph to a skilled photographer who has never seen the scene. The subject is the literal thing in the frame — a woman, a cathedral, a vintage motorcycle. The medium specifies how the image was captured or rendered: 35mm film photograph, oil painting, isometric 3D render, charcoal sketch. Style anchors the aesthetic and is where you reference an artist, era, or genre: in the style of Saul Leiter, brutalist architectural photography, 1970s sci-fi book cover. Lighting carries enormous weight in v7 because the model now simulates light physics more accurately — golden hour rim light, harsh midday sun, soft north-facing studio window, neon street reflections after rain. Composition tells the model how to frame the shot: wide establishing shot, dutch angle close-up, overhead flat lay, rule of thirds with negative space on the left. Finally, mood gives the model an emotional target: melancholic, triumphant, eerie, serene. A complete prompt might read: a lone fisherman repairing nets, weathered hands in focus, 35mm Kodak Portra 400 photograph, in the style of Steve McCurry, golden hour side light through a doorway, medium close-up with shallow depth of field, quiet and contemplative mood --ar 3:2 --style raw. The more of these layers you specify, the less the model has to guess.

Essential Parameters with Exact Syntax

Parameters go at the end of your prompt, each prefixed with two dashes. Here are the ones that matter in v7. --ar sets the aspect ratio: --ar 16:9 for cinematic widescreen, --ar 9:16 for vertical phone wallpapers, --ar 1:1 for square (the default), --ar 3:2 for classic 35mm, --ar 2:3 for portrait orientation. --v 7 forces the v7 model explicitly, though it's now the default; --v 6.1 or --niji 6 switch you back to older models or anime mode. --style raw turns down Midjourney's default aesthetic bias and produces more literal, photographic results — essential for product shots and reference photography. --stylize (or --s) controls how much artistic license the model takes, ranging from 0 to 1000 with a default of 100; lower values like --s 25 keep things grounded and faithful to your prompt, while --s 750 lets the model add dramatic flourish. --chaos (or --c) from 0 to 100 controls how varied the four output tiles are; --c 0 gives near-identical variations, --c 50 explores wider, --c 100 throws everything at the wall. --weird (or --w) from 0 to 3000 pushes the model toward unusual aesthetic choices and is excellent for concept art and surreal work. --no removes elements: --no people, --no text, --no blur. --tile makes the output a seamless repeating pattern, perfect for textures and wallpaper. --seed [number] locks the random seed so you can reproduce or carefully iterate on a specific generation. Example combining several: cyberpunk alleyway, neon signage in Japanese, light rain --ar 21:9 --style raw --s 200 --c 15 --no people.

Style Reference (--sref) and Niji 6 Workflows

Style Reference is the v7 feature that changes professional workflows. Instead of describing a style in words, you point Midjourney at one or more reference images and it extracts the aesthetic — color palette, lighting, texture, brush quality, grain — without copying the content. The syntax is --sref [image URL], and you can chain multiple URLs separated by spaces to blend styles: --sref https://example.com/img1.jpg https://example.com/img2.jpg. Add --sw (style weight) from 0 to 1000 to control how strongly the reference dominates; --sw 100 is the default, --sw 50 keeps your text prompt in charge, --sw 400 lets the reference take over. A common professional workflow is to build a 'style library' — three to five reference images that represent your brand or project aesthetic — and reuse them across every generation for visual consistency. Niji 6 is the dedicated anime and illustration model, activated with --niji 6 instead of --v 7. It understands Japanese illustration tropes far better than the main model: cel shading, manga linework, shounen action poses, ghibli-style landscapes. Mix and match by using --sref with --niji 6 to apply a specific illustration style across a series.

Character Reference (--cref) and Image Weights (--iw)

Character consistency was the single biggest weakness of earlier Midjourney versions. v7's --cref parameter solves this for most use cases. Provide a URL pointing to a clear, well-lit reference image of your character — ideally a portrait with the face fully visible — and Midjourney will preserve their features across new scenes. Syntax: a knight resting in a forest clearing --cref https://example.com/character.jpg --ar 16:9. Pair it with --cw (character weight) from 0 to 100 to control how much is preserved. --cw 100 (the default) preserves face, hair, and clothing; --cw 50 preserves face only, letting you change outfits; --cw 0 preserves face shape loosely for stylized reinterpretations. For comic panels, storyboards, or product photography with a recurring person, --cref is now reliable enough to ship client work. --iw (image weight) is a separate but related parameter that controls how much an input image influences the final output when you use image-to-image generation. With an image prompt in front of your text, --iw 0.5 makes the image a gentle suggestion, --iw 1 (default) balances image and text, and --iw 2 lets the image dominate while text only adjusts details.

Multi-Prompts and Weight Syntax

Multi-prompts let you explicitly tell Midjourney that your prompt contains multiple concepts and how much each one matters. The separator is :: (two colons). a forest::2 cyberpunk city::1 tells the model the scene is twice as much forest as cyberpunk city, producing a forest with cyberpunk elements bleeding through rather than the reverse. Without the weights, Midjourney would weight both concepts equally. Negative weights are extraordinarily useful: bright sunny meadow::1 flowers::-0.5 will produce a meadow noticeably emptier of flowers than the default. You can use multi-prompts to fix common interpretation issues — if Midjourney keeps adding extra people to a 'crowd watching fireworks' prompt and you only want a couple of figures, try crowd watching fireworks::1 large group::-0.8. Multi-prompts also work with style descriptors: oil painting::2 photograph::1 produces an oil painting with photographic precision rather than an averaged blur of both. Use them sparingly — overusing weights can make outputs feel disjointed. Two or three weighted concepts is usually enough.

Negative Prompting with --no

The --no parameter is the simplest and most reliable way to exclude unwanted elements. Syntax: --no [comma-separated list]. Example: serene mountain lake at dawn --no people, boats, buildings, text. Behind the scenes, --no applies a negative weight of -0.5 to each term, which is enough to suppress most common elements without producing the 'pink elephant' problem where mentioning something accidentally summons it. Use --no for predictable removals: text and watermarks (--no text, watermark, signature), unwanted body parts (--no extra fingers, deformed hands), clutter (--no background objects, distractions), or stylistic tendencies you want to avoid (--no cartoon, blur, low quality). For stronger removal, fall back to multi-prompts with explicit negative weights: --no isn't aggressive enough for stubborn elements, so concept::-1.5 in the main prompt body can do more. A common mistake is over-stuffing --no with vague terms like 'bad' or 'ugly' — these don't map cleanly to anything in Midjourney's latent space and waste prompt budget. Stick to concrete, removable nouns and styles.

Iteration Workflow: Vary, Pan, Zoom, and Upscale

Generating an image is only the first step. v7's iteration tools let you refine and extend any output. Once a grid finishes, click U1 through U4 to upscale one of the four variations to its full resolution. Below the upscaled image you'll find Vary (Subtle) for small refinements, Vary (Strong) for larger changes to the same composition, and Vary (Region) which opens an inpainting interface where you can paint a mask and rewrite just that area — invaluable for fixing a problem hand without rerolling the entire image. Zoom Out 1.5x and Zoom Out 2x extend the canvas outward, generating new content around your existing image; this is how you turn a portrait into a wide environment shot. Pan Up, Down, Left, and Right extend the image in one direction, useful for building panoramas or correcting awkward crops. Finally, Remix Mode (toggle with /settings) lets you change the prompt between iterations: upscale an image, then click Vary and edit the prompt to shift the season, time of day, or subject's expression while preserving the composition. The professional workflow is: draft mode to explore composition, full render to lock in style, Vary Region to fix details, Zoom Out to extend, and a final Upscale Subtle pass for the deliverable.

Common Pitfalls, Pro Tips, and Midjourney vs Flux 1.1 / SDXL

The most common Midjourney mistake is over-prompting. v7 responds to clarity, not word count. A 12-word prompt with strong nouns and one good lighting descriptor will usually beat a 60-word adjective soup. Cut filler words like 'beautiful', 'amazing', 'high quality', and '8k' — they signal nothing useful and were never necessary even in earlier versions. Instead, get specific: replace 'beautiful lighting' with 'soft window light from the left' and replace 'amazing detail' with the actual details you want. Second pitfall: ignoring --style raw for photographic work. Midjourney's default aesthetic adds a subtle painterly quality even to photo prompts; --style raw turns this off and is essential for product shots, reference imagery, and any work that needs to look like an unedited camera capture. Third pitfall: chasing seeds. Locking a --seed only reproduces an image if the prompt and parameters are identical; change one word and the seed becomes useless. Use seeds for fine-tuning, not for consistency — that's what --cref and --sref are for. How does v7 compare to alternatives? Flux 1.1 Pro still beats Midjourney on photorealism of human skin and on accurate text rendering, and its API is more developer-friendly. Stable Diffusion XL with the right LoRAs beats both on customization and offline workflows, but requires technical setup. Midjourney's edge is aesthetic consistency: out of the box, v7 produces images that look deliberately composed, with strong color, lighting, and design sensibility. For client work where you need polished output fast without fine-tuning models, v7 is still the most efficient tool in the stack.

Frequently Asked Questions