Midjourney drops V7 in alpha with new architecture and Draft Mode

Midjourney just dropped V7 into alpha testing, marking its first brand-new image model in nearly a year. The release landed right after OpenAI had everyone turning their photos into Studio Ghibli frames inside ChatGPT.

CEO David Holz described V7 as having a totally different architecture under the hood. The new model promises noticeably better textures, smarter comprehension of text prompts, and far better coherence on tricky details like hands, bodies, and everyday objects.

Expect a steady drip of updates every week or two over the next couple of months as the team irons things out.

Forced tastes and a 10x draft engine

To take V7 for a spin, you will encounter a brand new requirement: personalization is on by default.

If you have not already built a profile, you have to sit down and rate roughly 200 image pairs to teach the system your visual tastes. Once that is set up, you can toggle V7 on Midjourney's web platform or via its familiar Discord bot.

Because the underlying architecture has completely changed, old prompt formulas might not work the same way. A few staple features are also temporarily missing from the alpha, including image upscaling and retexturing.

Alongside the standard Relax and costlier Turbo modes, Midjourney introduced a new Draft Mode. It renders low-resolution drafts at ten times the speed and half the standard GPU cost. When you spot a keeper, you can upscale and re-render it in full fidelity with a single click.

Why it matters

Midjourney remains one of the weirdest and most fascinating outliers in Silicon Valley. Founded by Holz, the company has famously never raised a dime of venture capital, yet it reportedly pulled in roughly $200 million in revenue in late 2023, even while juggling copyright lawsuits from artists.

For creators and power users, Draft Mode alone justifies trying V7 right now. The biggest bottleneck in generative art has always been latency during the ideation phase; cutting that loop down to a tenth of the time changes how people brainstorm visual concepts.

However, forcing every user through a 200-image calibration test is a risky move that could alienate casual experimenters.

We think this release signals a smart defensive pivot. With OpenAI turning ChatGPT into an ultra-accessible consumer playground, Midjourney cannot rely solely on raw aesthetic prestige anymore. By overhauling its core architecture and building rapid drafting workflows, it is doubling down on professional creators who care far more about precision, speed, and iteration.