Midjourney has announced the alpha release of its V7 image generation model, now available for testing by the AI community. This new version boasts significant enhancements in text prompt comprehension, image quality, and the coherence of generated features. According to Midjourney, V7 displays remarkable improvements in understanding user prompts, producing stunning images with exquisite textures. Details in bodies, hands, and various objects show a marked improvement in coherence compared to previous models.
A standout feature of V7 is the model personalization function, which users can unlock in approximately five minutes. This feature, which can be toggled on or off at any time, is designed to enhance the AI’s understanding of individual user preferences and aesthetic tastes. Midjourney believes that this capability establishes a new standard for interpreting user intent. Additionally, Midjourney has introduced ‘Draft Mode’ alongside the V7 model, which allows for image generation that is ten times faster and at half the cost.
This speed enhancement has led to the incorporation of a new “conversational mode” in the web interface. In this mode, users can command the AI to make adjustments—like changing a cat to an owl or setting the scene to nighttime—and the system will automatically modify the prompt to create a new image. Draft Mode also includes voice input functionality, allowing users to verbally express their ideas and see images generated in near real-time. While draft images are of lower quality compared to those from the standard mode, they maintain consistent behaviors and aesthetic traits.
The V7 model will initially offer two speed modes: Turbo and Relax, with the standard mode undergoing optimization. Turbo jobs will cost twice as much, while draft jobs will be priced at half. Midjourney is also updating functionalities like upscaling and retexturing to the V6 model for the time being, with plans for future enhancements. Looking ahead, Midjourney has a robust development schedule and will roll out new features every one to two weeks for the next couple of months.
Users can also anticipate a forthcoming character and object reference capability. The company encourages experimentation with V7, reassuring users that it introduces new strengths and potential weaknesses, requiring different prompting techniques than earlier versions.