MidjourneyUnveilsV7A
作者：大自然的搬运工 · 2026-06-17 16:35:55
Midjourney has launched the alpha version of its V7 model, marking the AI image generator's first major architecture update in nearly a year. The release introduces a significant shift toward personalization, requiring users to build a profile by rating images before the system can generate outputs tailored to their aesthetic preferences. The standout feature is a default "Personalization" mode that learns individual taste, moving away from the generic "house style" that previously dominated the platform. Alongside this, V7 debuts "Draft Mode," which renders images at 10x the standard speed for rapid iteration, though at a lower quality. The company notes that current limitations include the absence of image upscaling, repainting, and standard aspect ratios beyond 1:1, with these features slated for weekly updates. Under the hood, V7 is a fully re-engineered model trained on a unified codebase that merges text and image prompts fluidly. Early testing suggests a dramatic improvement in handling complex prompts involving hands, bodies, and natural lighting, with a notably grittier and less synthetic aesthetic. The model also introduces "Sref Random" for exploring unexpected visual styles. Midjourney emphasizes that V7 is a "totally different architecture" designed to be smarter with text prompts, encouraging users to rely less on technical jargon and write more conversationally to achieve precise results.