What Midjourney v7 Actually Delivers
Midjourney v7, released in 2026, closed the gap between AI-generated and professional photography more than any previous version. Key improvements: dramatically better anatomy (no more morphing hands or duplicated fingers in the majority of outputs), consistent faces across multiple images in the same style, and a major leap in text rendering accuracy — the feature that had been AI image generation’s longest-running joke finally works reliably.
The tool remains Discord-based, though a standalone web app at midjourney.com now offers a full alternative to Discord for those who prefer not to navigate the Discord interface. Both modes produce the same quality output.
The Basic Prompt Structure
Midjourney prompts work differently from natural-language chatbots. The structure that produces the best results: describe the subject, then the setting or context, then the lighting and color mood, then the style reference, and finally any technical parameters.
Strong prompt example: “a samurai standing in fog-covered Kyoto alley at dawn, cinematic lighting, film grain, Fujifilm Provia 100f aesthetic, –ar 16:9 –s 200 –style raw”
Weak prompt example: ” samurai in Japan pretty”
The difference in output quality is substantial. Midjourney v7 rewards specificity. Describe what you actually want to see — not a general idea of it.
Key Parameters and What They Do
The –stylize parameter controls how artistic the output is, on a scale from 0 to 1000. Lower values produce more literal prompt-following; higher values produce more artistic, dramatic interpretations. The default is 100. For photorealistic work, values around 50 to 100 tend to work best. For editorial and artistic work, 200 to 400 produces more striking results.
The –chaos parameter controls variation — how different the four generated images are from each other. High chaos (60 to 100) produces wildly different options, useful for brainstorming. Low chaos (0 to 20) produces more consistent variations on the same interpretation, useful when you already know the direction and want to refine it.
The –ar parameter sets aspect ratio. –ar 16:9 for landscape web content, –ar 9:16 for mobile vertical, –ar 1:1 for square. Custom ratios like –ar 21:9 for ultra-wide cinematic are also supported.
The –style raw flag disables Midjourney’s automatic beautification, producing more literal prompt-following. For product photography and architectural visualization, –style raw consistently outperforms default settings.
Style References: Consistent Visual Identity
The most powerful new feature in recent versions: style references (–sref) and character references (–cref). Upload an image and Midjourney extracts its visual style — lighting, color grading, composition, texture — and applies that style to your new generation. This is the feature that makes Midjourney genuinely useful for brand-consistent content production.
For a brand that wants all its visuals in a specific look: generate a reference image that establishes the house style once, then use –sref with that image for every subsequent generation. The consistency is meaningful — not perfect, but good enough for content that does not need custom photography.
Omni Reference: Characters and Products
Omni Reference (–omni-ref) extends character consistency across sessions. Upload reference images of a product, character, or object, and Midjourney places that exact item in new scenes with consistent accuracy. For product marketing where the same product needs to appear in multiple contexts — lifestyle shots, seasonal campaigns, different platforms — this eliminates the need to reshoot or use composite workarounds.
The Subscription Tiers
The $10 per month Basic plan includes 200 fast GPU minutes — approximately 160 to 200 image generations at standard settings. This is enough for regular personal use but insufficient for professional volume. The $30 per month Standard plan adds 15 fast GPU hours plus unlimited relaxed-mode generation for queuing when you do not need immediate results. The $60 per month Pro plan adds stealth mode (private generations not visible in the community feed) and 30 fast GPU hours. The $120 per month Mega plan targets teams running high-volume production pipelines.
The Workflow That Actually Works
Start with –stylize 100 –chaos 30 for your first generation — enough variation to evaluate directions, enough consistency to recognize if the output is on target. Once you have a direction, reduce –chaos to 10 to 20 and increase –stylize to match the style you want. Use –style raw for photorealistic work. Use style references to lock visual identity once you have established it. Generate four images, upscale the best option, and use V (variation) on that upscale to iterate toward the final piece.