Midjourney has released an alpha version of its AI-powered image generator V7. Unlike OpenAI ChatGPT, this model has not been optimized to create Ghibli-style images, but that doesn’t stop it from generating aesthetically pleasing work, TechCrunch writes.
Image source: x.com/midjourney
Before you can start using Midjourney V7, you’ll have to rate about two hundred images so that the system can create a “personalization” profile for you, which helps customize the model to your visual preferences; this is the first time personalization has been enabled by default. Once you’ve set it up, you can start using V7 on both the Midjourney website and the company’s Discord server. It was developed using a “completely different architecture,” said Midjourney CEO David Holz.
The model is available in two variants: the more resource-intensive Turbo; and Relax with Draft Mode, which generates images ten times faster and uses half as much computing power. Draft Mode “draft” images are of lower quality than those created in Standard mode, but they can be improved and re-rendered with a single click. The model has its advantages and disadvantages, Mr. Holz warned, so users are advised to experiment with querying to achieve the best results.
Midjourney is an unusual company. Founded in 2022 by David Holtz, a one-time co-founder of peripherals maker Leap Motion, the startup is notable for its complete lack of outside funding. In late 2023, the company said it expected to make about $200 million in revenue. Last year, Midjourney announced it was hiring for a hardware project, while continuing to train previously announced models for generating video and 3D objects.