Midjourney V7
Midjourney V7: New Architecture, Sharper Coherence, Draft Mode
Midjourney V7 is the first major Midjourney upgrade in nearly a year — released as Alpha on April 3, 2025 and built on a brand-new architecture rather than an evolution of V6. The result is noticeably sharper image coherence, stronger prompt adherence, more believable hands and anatomy, and the painterly atmosphere Midjourney is known for. V7 also introduces Draft Mode (roughly 10× faster, lower cost, lower fidelity for ideation), Omni-Reference (--oref) which generalises Midjourney's old character-reference to any subject — people, objects, vehicles, characters — and Style Reference (--sref) for consistent visual identity across a series. LoveGen AI exposes V7 with three render speeds (draft / fast / turbo), four unique candidates per task, and the full native flag syntax — no Discord required.
V7 is the most significant Midjourney release since V6 (December 2023). The team rebuilt the model from scratch instead of iterating, focusing on three problem areas creators reported most often. First, prompt adherence: V7 reads prompts more literally, so multi-clause descriptions with specific spatial and material instructions resolve closer to what you wrote. Second, fine-detail coherence: hands, fingers, hair strands, fabric folds, jewellery, and small props now hold up under closer inspection — a long-running pain point in earlier versions. Third, light and material rendering: skin, metal, glass, water, and fabric all read as the right substance, with V7 keeping Midjourney's signature cinematic light and atmosphere.
Three V7-era workflow features change how you use the model day-to-day. Draft Mode renders roughly 10× faster than Standard at lower fidelity and lower cost — designed for rapid ideation rather than final delivery; once a draft composition lands, you can re-render it at fast or turbo for production output. Omni-Reference (--oref) is V7's evolution of character reference: instead of being limited to faces, it locks any subject — a specific bottle of wine, a sneaker, a car silhouette, a costumed character — across multiple generations, useful for product photography and consistent series art. Style Reference (--sref) carries colour, mood, lighting, and rendering style from one image to another without copying the subject. All standard parameters return as well: --ar (aspect ratio), --s (stylize, 0–1000), --c (chaos, 0–100), --weird (0–3000), and --iw (image weight, 0–3) for image-to-image control.
On LoveGen AI, V7 runs through Midjourney's official inference partner. The model is locked to V7 (--v / --version are stripped), --niji is not supported, and the speed flags (--draft / --fast / --turbo) are controlled by the Speed selector rather than embedded in your prompt. Every task returns four unique candidates rendered as a 2×2 grid; click any thumbnail to view full-size or download. Midjourney's content policy is enforced server-side: prompts that violate policy are filtered and credits for filtered tasks are not refunded, so prompts should stay within Midjourney's published Community Guidelines.
How to Use Midjourney V7 on LoveGen AI
Write a Specific, Art-Directed Prompt
MidjourneyV7Page.howToUse.steps.0.description
Choose Aspect Ratio, Speed and (Optional) References
Pick 1:1, 16:9, 9:16, 4:3 or 3:4 — the aspect ratio is appended as --ar automatically. Pick Draft to brainstorm cheaply, Fast for production, or Turbo for priority queue. Add up to 4 reference images for image-to-image: a single image must be paired with text, two or more can run with or without text.
Generate, Compare 4 Candidates, Download
Each task returns four unique images as a 2×2 grid. Click any thumbnail to enlarge, hover to download a single image, or use the Download button to save the full set. If a draft composition is right, re-run the same prompt at Fast or Turbo for the production render.
Midjourney V7 Technical Specifications
| Provider | Midjourney, Inc. |
| Released | Alpha — April 3, 2025 |
| Model ID | mj-v7 |
| Architecture | Ground-up new model (not iterated from V6) |
| Images per Task | 4 unique candidates |
| Speed Modes | Draft (10× faster) / Fast (default) / Turbo (priority) |
| Aspect Ratios | Any --ar — UI presets: 1:1, 16:9, 9:16, 4:3, 3:4 |
| Stylize (--s) | 0–1000 (default ~100) |
| Chaos (--c) | 0–100 (variation between the 4 candidates) |
| Weird (--weird) | 0–3000 (unconventional aesthetic) |
| Image Weight (--iw) | 0–3 (default 1) |
| Omni-Reference | --oref — lock any subject (people, objects, products) |
| Style Reference | --sref — transfer colour, mood, rendering style |
| Image-to-Image | Reference URL(s) prefixed in prompt — 1 image needs text, 2+ images optional text |
| Prompt Length | Up to 8,192 characters |
| Approx. Latency | Draft ~30s · Fast 1–3 min · Turbo seconds–1 min · Hard cap 20 min |
| Locked Flags | --v / --version (locked to V7), --niji (unsupported), --draft / --fast / --turbo (use Speed selector) |
Why Choose Midjourney V7?
Brand-New V7 Architecture
V7 is the first ground-up Midjourney rebuild since V6 — not an iteration. Sharper coherence, more believable hands and anatomy, more literal prompt adherence, and Midjourney's signature cinematic light and atmosphere preserved.
Draft Mode for Real Iteration
Draft renders roughly 10× faster than Standard at a fraction of the cost. Generate dozens of compositions in the time one Fast render takes, then re-run the winners at Fast or Turbo for final fidelity.
Omni-Reference: Lock Any Subject
Omni-Reference (--oref) generalises Midjourney's character reference to any subject — people, products, vehicles, costumes — keeping the same identity across multiple scenes. Critical for product shoots and consistent series work.
Style Reference for Visual Identity
Style Reference (--sref) transfers colour palette, lighting, and rendering style from one image to another without copying the subject. Use it to keep a campaign or brand series visually consistent.
Full Native Flag Syntax
Same vocabulary as Discord and the Midjourney web app — --ar, --s (0–1000), --c (0–100), --weird (0–3000), --iw (0–3). No separate UI to relearn, paste a known prompt and it works.
Four Candidates — No Re-Rolls
Every task returns four unique compositions. You compare and pick the strongest instead of re-rolling, converging on the right shot in fewer credits than single-output models.
Midjourney V7 vs Other AI Image Generators
| Feature | Midjourney V7 | GPT Image 2 | Flux 2 Pro | Nano Banana Pro |
|---|---|---|---|---|
| Provider | Midjourney | OpenAI | Black Forest Labs | |
| Released | April 2025 (Alpha) | April 2026 | November 2025 | January 2026 |
| Best For | Aesthetic, cinematic, art direction | Multilingual text & reasoning-led layouts | Studio-grade photorealism | Natural-language image editing |
| Candidates per Task | 4 unique | 1 | 1 | 1 |
| Speed Tiers | Draft / Fast / Turbo | Single tier | Single tier | Single tier |
| Subject Reference | Omni-Reference (--oref) | Up to 4 reference images | Up to 8 reference images | Up to 14 reference images |
| Style Reference | Yes (--sref) | Implicit via reference | Implicit via reference | Implicit via reference |
| Native Flag Syntax | Full (--ar / --s / --c / --weird / --iw / --oref / --sref) | No | No | No |
| Text Rendering | Limited | 99%+ across 6 languages | Excellent (English) | Excellent |
| Strongest Use Case | Concept art, posters, mood-led photography | Multilingual ads, infographics, UI mockups | Editorial photo realism | Product editing & consistency |
MidjourneyV7Page.useCase.h2
Concept Art & Pre-Production
Build mood, key art, and visual development for film, games, and animation. Draft Mode lets a director or art lead burn through hundreds of compositions in a single afternoon.
Brand & Editorial Photography
Produce hero imagery with Midjourney's signature cinematic lighting. Use Style Reference (--sref) to lock a campaign look and Omni-Reference (--oref) to keep a product or model consistent across the shoot.
Album, Book & Magazine Covers
Four candidates per prompt means four cover ideas in one run. Tune --c to widen variety, --s to push stylization, --weird to break out of safe choices.
Character Design & Style Sheets
Lock a character's identity with Omni-Reference, then re-pose, re-light, and re-cost them across scenes. Useful for tabletop RPG art, comic pre-production, and animation development.
Print, Posters & Merchandise
Output at print-friendly aspect ratios (3:4, 4:3) and up-rez offline. V7's improved coherence means fewer hand-fix and detail-touch-up passes before press.
Architectural & Product Visualisation
Space, depth, materials, and reflections all read more believably in V7. Use it for early-stage interiors and exteriors, or to render a product hero shot from a clay reference.
Explore Other AI Image Models
GPT Image 2
OpenAI's reasoning-first generator with 99%+ multilingual text accuracy at 2K.
Nano Banana Pro
Google's natural-language image editor with up to 14 reference images.
Flux 2 Pro
Black Forest Labs' studio-grade generator with 4MP output and multi-reference control.
Imagen 4
Google DeepMind's photorealistic image model with precise aspect-ratio control.
Ideogram v3
Typography-focused generator that nails logos, posters, and text-in-image designs.
Seedream 4
ByteDance's high-quality generator with precise aspect-ratio control and rich detail.
Frequently Asked Questions About Midjourney V7
What is Midjourney V7 and when was it released?
Midjourney V7 is the seventh major version of the Midjourney image model, released as Alpha on April 3, 2025 — the first major Midjourney upgrade in nearly a year (V6.1 launched in August 2024). It's a brand-new architecture, not an iteration of V6.
How is V7 different from V6 / V6.1?
V7 is built from scratch, not derived from V6. The most visible improvements are: (1) noticeably better fine-detail coherence — hands, hair, fabric, jewellery and small props hold up under close inspection; (2) more literal prompt adherence; (3) Draft Mode for fast ideation; (4) Omni-Reference (--oref) which generalises character reference to any subject; and (5) refined Style Reference (--sref). The painterly atmosphere and cinematic light Midjourney is known for are preserved.
What is Draft Mode?
Draft Mode renders roughly 10× faster than Standard at lower fidelity and lower cost. It's designed for rapid ideation — generating many compositions cheaply — not for final delivery. When a draft composition lands, re-run the same prompt at Fast or Turbo to produce a higher-fidelity render.
What's the difference between Draft, Fast and Turbo?
Draft is the cheapest and fastest mode (lower fidelity, ideation use). Fast is the production default. Turbo runs on a priority queue at higher cost — use it when you need a result immediately. On LoveGen AI you select the mode in the Speed selector; the --draft / --fast / --turbo flags inside a prompt are stripped server-side.
What is Omni-Reference (--oref)?
Omni-Reference is V7's evolution of character reference. Instead of only locking faces or characters, it can lock any subject — a specific product, vehicle, costume, or person — across multiple generations. Useful for product photography, consistent series art, and brand assets.
What native parameters / flags work with V7?
All standard Midjourney V7 parameters work as inline prompt flags: --ar (aspect ratio), --s (stylize, 0–1000), --c (chaos, 0–100), --weird (0–3000), --iw (image weight, 0–3), --sref (style reference URL), and --oref (omni-reference URL). The model is locked to V7 — --v / --version are stripped — and --niji is not supported.
Can I do image-to-image?
Yes. Upload up to 4 reference images and they're prefixed to your prompt as URLs. Midjourney's input rules apply: a single reference image must be paired with a text description; two or more images can run with or without additional text.
Why does each task return four images?
Midjourney has always returned four unique candidates per task so you can compare compositions and pick the strongest result — closer to a contact sheet than a single render. LoveGen AI shows the four as a 2×2 grid; click any thumbnail to expand.
What happens if my prompt is filtered by content moderation?
Midjourney enforces its content policy server-side. Tasks blocked by moderation do not return images, and per Midjourney's policy credits for filtered tasks are not refunded. Keep prompts within Midjourney's published Community Guidelines.