Midjourney just put a deadline on its next model. At the developer office hours on November 11, David Holz said V9 should ship within 30 days, possibly sooner, and that a separate preview run is not planned. The base model is already good enough to release; what is left is bundling the rest of the launch around it.
If Midjourney is part of your work, that is a scheduling problem rather than a curiosity. There is no long preview window to compare old and new outputs side by side and migrate at your own pace. The default model flips, and every prompt you have tuned since V8.2 became the default in July starts producing slightly different images.
What V9 Actually Changes
The launch is not a single feature. It pairs a new base model with a rebuilt upscaler, plus a new SREF system that is still unfinished. Midjourney says consistency, prompt adherence and moodboards all improve, with finer and more creative detail in the output. Default aesthetics and personalization are expected to stay where they are, so a profile you have spent a year tuning should not need rebuilding from scratch.
Text rendering is the loudest goal. Early tests handle fairly long passages, though not walls of copy, and edge cases are expected at launch. That matters for mockups, packaging concepts and anything with signage in frame, where V8 still needs manual cleanup.
A Thinking mode is coming too. It improves prompt adherence while slightly hurting aesthetics today, so the team is tuning it to work as a default rather than an option. A stronger version exists at roughly ten times the cost.
Real-time generation is the piece nobody should plan around. Holz likes two prototypes, a third is at an earlier stage, and the intended flow is to iterate fast at low quality and then render the final frame at maximum quality. How GPU time gets split between those two passes has not been decided.
Money is the other open question. V9 images could cost double to generate. Nothing is final, but Holz said he would rather take a large jump in quality than shave compute to hold the price.
- Text in images becomes usable. Short passages of legible type stop being a coin flip, so packaging and poster concepts need fewer fixes.
- Saved SREF codes are at risk. A new reference system means codes tuned on V8.2 may read differently, and you want the old results saved before that happens.
- Cost per usable image may climb. If the double-cost estimate lands, plan fewer exploratory rerolls and more deliberate prompting.
- Thinking mode rewards specificity. If adherence improves, vague mood prompts lose their charm and structured descriptions start winning.
Your 30-Day Prep Checklist
You cannot move the launch date, but you can control how much of your pipeline breaks on it. This fits into a few evenings.
- Freeze a reference set. Choose 10 to 15 prompts that represent your real work and save the outputs now with the version pinned, so you have a baseline to compare against.
- Pin the version everywhere. Keep an explicit version flag in shared prompts and templates until launch day. Silent default changes are how a team ends up with inconsistent assets inside one campaign.
- Export styles and moodboards. Copy SREF codes, style references and moodboard contents into a document outside Midjourney. When the new reference system ships, you will want to know what the old ones looked like.
- Rerun the reference set first. On launch day, batch your baseline prompts before touching paid client work. A 20 minute comparison tells you whether the new model is an upgrade or a recalibration for your aesthetic.
Keep a Legacy Escape Hatch
The website changes at the same time as the model. Some alpha features move to the main site, some get rolled back, and the current site is expected to stay online as a legacy version for V8.2 workflows. The new site will most likely become V9's home, though no date has been given.
- Bookmark the legacy site now. Do not rely on finding it later from a redesigned product page you have never seen.
- Write down your settings. Default style, personalization state, upscale preferences and any per-job flags belong in your own notes, not only in a profile you do not control.
- Archive your best work. Keep the full resolution files and the prompts that produced them, in your own storage, before the interface moves underneath you.
One more date for the calendar. Niji is in testing with guides and moderators and could land within a week, separately from V9. Midjourney also says Niji and V9 are the last major model releases this year, with V9.1 pencilled in for next year and V10 plus video generation still in planning, possibly with a 3D step in between. A dedicated image-creation device is in early development with no release date.
None of this requires you to change how you prompt today. It requires you to have evidence of how you prompt today, which is the part most people skip and then regret. Pin the version, save the references, and keep a baseline you can rerun in half an hour.
When V9 flips on, the creators who will notice the least are the ones who already have next month's comparisons sitting in a folder. The launch is 30 days away at most. That is plenty of time to be ready, and no time at all to start from scratch.
Comments