What Is Midjourney V7 and Why Does It Change How You Work?
Most people using Midjourney V7 are still prompting the same way they did in V5. Same vague descriptions, same settings, same trial-and-error loops — and they're wondering why V7 doesn't feel dramatically different. The answer is that V7's biggest improvements aren't automatic. They require setup. Draft Mode, model personalization, and Omni Reference are all features you have to deliberately activate. Once you do, the workflow changes significantly.
Midjourney V7 is the seventh major version of the platform's image generation model, released in alpha in early 2026. It is the first Midjourney model to turn on model personalization by default, the first to introduce Omni Reference for cross-image consistency, and the first to include native video generation — clips up to 21 seconds. The underlying rendering engine also produces markedly better coherence in complex scenes: hands, architectural details, and layered textures are all more reliable than in V6.1.
Early comparisons published by DataCamp and Tom's Guide show V7 outperforming V6.1 consistently on detailed product shots, portrait compositing, and cinematic stills. But those results assume users are prompting for V7 — not still using V5-era habits.
What Is Draft Mode and How Do You Use It to Save 60% of Your Credits?
Draft Mode is Midjourney V7's biggest workflow upgrade. It generates images at 10 times the speed and half the GPU cost of standard mode. This makes it the correct choice for the exploration phase of any project — and most practitioners are not using it yet.
Here is how Draft Mode changes the workflow in practice. On the Midjourney web interface, activating Draft Mode switches the prompt bar into a conversational mode. You describe what you want, see the result in seconds, and then refine it in plain language — tell it "make the lighting warmer" or "change the background to a rooftop at night" and it adjusts the prompt automatically and generates a new version. When you find a direction that works, click "Enhance" to re-render at full quality.
To activate Draft Mode on web: enable it in the Settings panel before entering your prompt. On Discord: add --draft to the end of your prompt (note that conversational mode is web-only). Draft images are deliberately lower quality than standard renders — never deliver them as finals. The correct workflow is 5–10 draft iterations to lock concept and composition, then one full Enhance render for delivery. According to Midjourney's documentation, this approach saves approximately 60–70% of GPU credits on exploration-heavy projects compared to running standard renders at every step.
How Does Model Personalization Work in V7 — and Why You Need to Set It Up Now?
Model personalization means V7 learns your aesthetic preferences and adjusts its default outputs to match. It is turned on by default — which means if you have not set up your personalization profile, V7 is already applying a generic aesthetic instead of yours. Setting it up is a one-time process that takes about 15 minutes.
The setup process works as follows. Go to the Midjourney website and click the "Personalize" button in the sidebar. Midjourney shows you pairs of images and asks which you prefer. Rank at least 200 pairs to unlock personalization for V7 — most power users aim for 500 or more rankings to get meaningfully calibrated results. Your personalization profile is tied to your account and applies automatically to every V7 generation.
What changes after personalization: if you consistently click editorial photography over soft illustration, V7 begins defaulting toward harder lighting, stronger shadows, and more editorial framing — even when you do not specify these in the prompt. The prompt provides direction; personalization handles the aesthetic finishing. To disable personalization for a specific prompt when you need a neutral output for client work, add --no personalization to your prompt.
What Is Omni Reference and How Do You Use It for Consistent Results Across Multiple Images?
Omni Reference (--oref) is V7's solution to one of the most persistent problems in AI image generation: visual consistency across multiple outputs. If you need the same product, character, environment, or visual style to appear recognizably across a series of images, Omni Reference lets you anchor a reference image and control how closely V7 adheres to it.
How to use Omni Reference: upload your reference image to Midjourney, copy its URL, and add --oref [IMAGE_URL] --ow [0–100] to your prompt. The --ow parameter is the Omni Weight — 0 means loose reference (mood and color only), 100 means tight reference (close visual match). For a product shot series where the packaging must stay consistent across five different lifestyle scenes, use --ow 80–100. For borrowing a mood or color palette without replicating the exact form, use --ow 20–40.
Omni Reference differs from the older --cref (Character Reference) parameter, which only works with character faces and poses. Omni Reference works with objects, products, scenes, and architectural elements — making it the more versatile tool for commercial content workflows.
How Do You Write Better Prompts for V7 Compared to Earlier Versions?
V7's prompt engine is significantly more literal than V5 or V6. In earlier versions, vague prompts often produced interesting results through creative interpretation. In V7, the model follows your prompt more precisely — which means vague prompts produce average results instead of surprising ones. The practical shift is: be more specific, not less.
V7 rewards prompts built on a clear structure: [Subject] + [Action/Setting] + [Lighting] + [Style/Aesthetic] + [Camera Parameters] + [Technical Flags]. Style descriptors matter more than in V6: "cinematic editorial" produces different results from "magazine photo," and "documentary street photography" produces different results from "candid shot." Technical camera parameters — lens, aperture, focal length — now meaningfully change output in V7 in ways they did not in V5.
Example of a V7-optimized prompt versus a V5-era prompt. V5-era: "professional woman in an office, looking confident." V7-optimized: "Hong Kong female executive, 35–40, reviewing a data dashboard on a large monitor in a modern glass-walled office, diffused afternoon light from the left, editorial magazine style, 85mm portrait lens f/2.0, Kodak Portra 400 film grain --v 7 --ar 16:9 --s 750." The second version will produce a result that looks like a deliberate photographic choice, not a stock image.
What Are V7's New Film LUTs and Lens Control Parameters — and Why Do They Matter for Practitioners?
V7 introduced two parameters that give practitioners precise control over the visual language of generated images: Film LUTs and Lens Control. Both work through natural language descriptors in the prompt rather than sliders or settings panels.
Film LUTs apply cinematic color grading to your generated image as part of the generation itself — not as post-processing. V7 responds meaningfully to LUT-style language that cinematographers and photographers use. Examples that work in V7 prompts: "Kodak Portra 400 film emulation, warm highlights, slight grain" for a warm editorial look; "Fuji Velvia color saturation, rich greens and blues, high contrast" for nature and landscape work; "matte finish, lifted blacks, teal shadows, orange skin tones" for a modern commercial aesthetic; "high contrast black and white, deep shadows, Zone System exposure" for dramatic monochrome.
Lens Control simulates real camera optics through focal length and aperture descriptors. V7 responds accurately to: "24mm wide angle" for environmental context with edge distortion; "85mm portrait lens" for natural perspective with subject separation; "200mm telephoto" for compressed background depth; "f/1.2" for extreme shallow depth of field with prominent bokeh; "f/11" for sharp throughout the frame. Combining Film LUTs and Lens Control — "Kodak Portra film emulation, 85mm lens, f/2.0, golden hour light" — produces results that look remarkably close to an actual photographic workflow.
Where Does Midjourney V7 Still Fall Short?
V7 is a significant upgrade, but being clear about its limitations will save you credits and client frustration. Three areas where V7 still underperforms are worth knowing before you commit to a workflow that depends on them.
Text rendering: V7 has improved on V6 for text-in-image, but it still makes errors on longer strings and complex scripts — including Chinese characters. Use Midjourney for composition and visual concept; use Canva, Adobe Firefly's text tools, or Photoshop for final text overlays. Complex hand gestures: V7 handles static hands and standard poses well, but intricate gestures at unusual angles still show anatomical errors at a rate higher than facial features. The "too literal" problem: VentureBeat's coverage of the V7 launch documented a portion of the user community finding the model's increased literalness reduced creative surprise — practitioners who relied on V5/V6's creative interpretation will need to invest more effort in structured prompting to compensate.
Try This Prompt: A V7 Template You Can Copy and Use Right Now
Copy this template into Midjourney V7 and adjust the subject and setting for your use case. It combines all the V7 features covered in this article — personalization will activate automatically if you have set up your profile.
Commercial Lifestyle Shot (Product / Brand):
--- Hong Kong business professional, 30–38, confident and relaxed expression, working with a laptop and coffee in a bright contemporary co-working space, soft diffused daylight from large windows on the left, editorial commercial photography style, Kodak Portra 400 film emulation, 85mm lens f/2.0, shallow depth of field on background, no text or logo --v 7 --ar 16:9 --s 750 --q 2
What each element does:
--- Subject age range: avoids generic "stock photo" face generation
--- Setting specificity: "co-working space" reads more contemporary than "office"
--- Lighting direction: "left" creates natural shadow fall for dimensional depth
--- Film LUT: Portra 400 adds warmth and slight grain that reads as "real photo"
--- 85mm f/2.0: creates the subject separation you see in premium brand photography
--- --s 750: Stylize at 750 balances prompt adherence with aesthetic polish
--- --q 2: Full quality; swap for --draft during concept exploration
Run this in Draft Mode first, iterate 4–5 times until the composition is right, then hit Enhance once for the final delivery-quality version.
Upgrade Your AI Workflow with UD
Knowing the tools is step one. Integrating them into a workflow that produces consistent, professional results — that's where most practitioners get stuck. UD 團隊手把手帶你完成每一步，from tool configuration and prompt systems to workflow design and deployment, so AI genuinely works for you every day.
懂AI，更懂你 — UD相伴，AI不冷。測試你的 AI 技術水平，看看你在哪個層級：