# V7 New Features / V7 新功能详解 > Complete guide to what's new in Midjourney V7 compared to V6. --- ## Overview V7 was released on **April 3, 2025** and became the default model on **June 17, 2025**. It is described as a "ground-up rework" with major improvements across the board. --- ## Major New Features ### 1. Video Generation | Detail | Info | |--------|------| | Max length | Up to 21 seconds | | Quality by length | 5s = excellent, 10s = good, 15s = decent, 21s = mixed | | Aspect ratios | Multiple supported | | Camera movements | Smooth, cinematic quality | V7 is the **first Midjourney model to support video generation**. Use camera movement keywords in your prompts: | Keyword | Effect | |---------|--------| | `FPV flight` | First-person drone racing | | `Tracking shot` | Camera follows subject | | `Orbital` | 360-degree rotation | | `Crane down` | Descending aerial | | `Push-in` | Moving closer slowly | | `Low angle` | Ground-level looking up | --- ### 2. Draft Mode - **10x faster** generation - **Half the GPU cost** - Lower resolution output for rapid prototyping - Can upscale and enhance later - Perfect for iterating on prompt ideas before committing to full-quality renders --- ### 3. Conversational Mode - Interactive, chat-like prompting workflow - Say "make it winter" and it updates the scene while keeping the same composition - Voice input supported - Enables iterative refinement without rewriting entire prompts --- ### 4. Personalization (`--p`) - **Turned on by default** for the first time in V7 - Requires rating ~200 images to build your aesthetic profile - Learns your visual preferences over time - Produces output more aligned with your taste --- ### 5. Omni Reference (`--oref`) V7-exclusive feature that replaces and extends V6's `--cref`: | Feature | Old (--cref) | New (--oref) | |---------|-------------|-------------| | Characters | Humans only | Characters, objects, vehicles, creatures | | Weight control | Limited | `--ow` parameter (community recommends ~400) | | Versatility | Face-only | Full-body and non-human subjects | | Compatibility | Standalone | Works with --p, --s, --sref, moodboards | Usage: ``` /imagine prompt text --oref [image URL] --ow 400 --v 7 ``` See the [Omni Reference Guide](../parameters/oref-omni-reference.md) for full details. --- ### 6. Experimental Parameter (`--exp`) - Range: 0-100 - Adds advanced tone mapping and detail enhancement - **Best kept under 25** for consistent results - Over 25 produces unpredictable but creative outputs - Community-discovered parameter, not officially documented See the [Experimental Parameter Guide](../parameters/exp-experimental.md) for full details. --- ### 7. Higher Quality Tier (`--q 4`) - New quality level beyond `--q 2` - Maximum refinement at higher GPU cost - Best for final renders where every detail matters --- ## Image Quality Improvements | Aspect | V6 | V7 | |--------|----|----| | Photorealism | Good, occasional artifacts | Exceptional, fewer issues | | Skin textures | Decent | Significantly improved | | Fabric details | Acceptable | Much more realistic | | Light and shadow | Good | Superior | | Composition | Good | Better balanced | | Hands/fingers | Problematic | Fewer weird artifacts | --- ## Prompt Understanding Improvements V7 is **"much smarter with text prompts"** per community consensus: - Understands natural language better than keyword stuffing - Knows what "golden hour" and "cyberpunk aesthetic" mean without spelling out every detail - Text rendering improved (still imperfect — keep to 2-4 words) - Context understanding significantly enhanced --- ## Access and Pricing | Plan | Price | Fast Hours | Key Features | |------|-------|-----------|-------------| | Basic | $10/month | ~200/month | Casual use | | Standard | $30/month | 15 hours | Unlimited relaxed mode | | Pro | $60/month | 30 hours | 12 concurrent jobs + stealth mode | | Mega | $120/month | 60 hours | Heavy users and professionals | All tiers get V7 access. Commercial use requires Pro subscription. --- ## V7 vs V6 Parameter Changes - `--stylize` range expanded from 750 (V6 max) to 1000 - `--chaos` behaves more predictably - `--weird` not adjustable in V7 web UI (produces errors) - Aspect ratios more flexible - **New parameters:** `--oref`, `--ow`, `--exp`, `--q 4` - Old V6 prompts may produce different results in V7 — retest them --- ## Next Steps - [Getting Started Guide](getting-started.md) — Basics of prompt writing - [Prompt Structure](prompt-structure.md) — Deep dive into prompt engineering - [Browse Prompts](../prompts/photography.md) — Ready-to-use examples - [Parameter Guides](../parameters/stylize-comparison.md) — Master every parameter