V8模型发布进展及策略

📅⛵ Recap Midjourney Office Hours January 21, 2026 In few words: The V8 model is entering its final phase and approaching release with a gradual launch strategy. Major work is underway on prompt management, text, customization, and the interface, with a potential release date in early February. 🇫🇷 Version française en commentaire 🔽 🔶 V8 Image Model Status & Timeline 🔹 Primary focus is preparing the V8 image model for launch. 🔹 Currently in internal testing (before guides/mods). 🔹 Model is in pre-optimization phase, so generation is slower than final. 🔹 Many features are still being connected due to: 🔸 New codebase and new APIs. 🔸 V8 being significantly different from prior models. 🔹 Guides and moderators are expected to start testing this week. 🔹 V8 will not release in January. 🔹 Earliest likely release window is early February, assuming things go smoothly. 🔹 Release timing depends on: 🔸 Bug fixes. 🔸 Feature readiness. 🔸 Optimization duration (could take ~1 week, uncertain). 🔶 V8 Release Strategy (Rolling / MVP Approach) 🔹 Team is debating a non-traditional rollout: 🔸 Possible early release before full training is complete. 🔸 Continue training and improving post-launch with frequent updates. 🔹 Open question on whether to wait for: 🔸 Upscaler. 🔸 Editing tools. 🔹 Likely outcome: 🔸 Launch with a minimal viable feature set, then add features incrementally. 🔹 Core challenge is deciding when V8 is “good enough” to ship. 🔶 Prompt Handling & Model Behavior Improvements 🔹 Major focus on making V8 work well across all prompting styles: 🔸 Short prompts. 🔸 Long prompts. 🔸 Mixed or unconventional formats. 🔹 Current issue: 🔸 Model is somewhat finicky depending on prompt style. 🔹 Goals: 🔸 Consistent quality regardless of prompt length or structure. 🔸 Avoid prompt rewriting by the model. 🔸 Strong out-of-the-box behavior without prompt engineering. 🔶 Bugs, Testing, and Optimization 🔹 At least 1–2 known bugs actively being addressed. 🔹 Process: 🔸 Fix bugs. 🔸 Retest. 🔸 Repeat until stable. 🔹 Feature hookup and bug iteration happening simultaneously. 🔹 Optimization phase still pending after features stabilize. 🔶 Editing Model & Resolution Plans 🔹 Separate edit model is in development alongside V8. 🔹 A smaller model is also being explored. 🔹 Model supports: 🔸 Multiple resolutions. 🔸 Multiple output sizes. 🔹 Still undecided: 🔸 Which resolutions map to which use cases. 🔹 Possible future plans: 🔸 Native 2048 resolution mode. 🔸 High-resolution training later (not started yet). 🔶 UI & Style Creator Updates 🔹 Ongoing UI improvements across the platform. 🔹 Style Creator updates: 🔸 New scrolling-based interface. 🔸 Ability to initialize using: ▪️ Style references (srefs). ▪️ Search terms. 🔹 Acknowledgement that: 🔸 Users still won’t be fully satisfied. 🔸 A larger third redesign is planned later. 🔹 Short-term plan: 🔸 Ship incremental improvements soon. 🔸 Accept tradeoffs rather than blocking progress. 🔶 Personalization System Overhaul 🔹 New scrolling-based personalization interface in development. 🔹 Similar interaction style to the updated Style Creator. 🔹 Key improvements: 🔸 No more clicking between two images. 🔸 Smoother and more intuitive flow. 🔹 Current personalization is widely disliked. 🔹 New version aims to be: 🔸 Better overall. 🔸 Not worse in any major way. 🔹 Team feels close to “flipping the switch” to the new system. 🔹 Ranking party may still be required before launch: 🔸 Improves quality. 🔸 Could delay release slightly. 🔶 Feature Load & Development Constraints 🔹 V8 includes far more than basic text-to-image: 🔸 Style references. 🔸 Mood boards. 🔸 Personalization. 🔸 Chaos / weird parameters. 🔸 Style creator. 🔸 Upscaling. 🔸 Editing. 🔹 Reality: 🔸 More features than people on the team. 🔹 Development effort is spread thin but progressing. 🔶 Backwards Compatibility & Niji Integration 🔹 Personalization and mood boards will be added to Niji G7. 🔹 Lack of these features at Niji G7 launch had no strong reason—just timing. 🔹 Focus on: 🔸 Backwards compatibility. 🔸 Avoiding unnecessary model splits. 🔶 Niji v7 Reception & Feedback 🔹 Niji v7 response has exceeded expectations. 🔹 Strong positive feedback from users. 🔹 Noted strengths: 🔸 Clean, confident line work. 🔸 Vibrant colors. 🔸 Improved coherence. 🔸 Better prompt following. 🔹 Text rendering improvements exist, but: 🔸 Text is not the primary goal of Niji. 🔸 V8 will handle text significantly better. 🔹 Team is actively listening for what users value most. 🔶 Text Rendering Outlook 🔹 V8 will have much stronger text performance than current models. 🔹 Users focused on text should expect major gains with V8 rather than Niji. 🔶 Overall Outlook 🔹 V8 is in the final stretch, but still has many moving parts. 🔹 Progress is steady, not blocked by a single issue. 🔹 Expect: 🔸 Early February release. 🔸 Possible rolling updates post-launch. 🔹 Team confidence is high; quality is already strong and improving rapidly.