Midjourney launches its new imaging tool, V6, which has also learned to create texts

The alpha version of Midjourney V6, an image-generating AI, has been released. Among the notable enhancements, developers highlight more realistic and detailed images, along with the model's ability to generate legible text within images.

Midjourney V6 is described as the "third model trained from scratch on our AI superclusters," and it took nine months to develop, explained company head David Holz. Connection to the updated neural network is not automatic for all users—users need to switch to the V6 mode in the Discord chat with the Midjourney bot using the "/settings" command and then selecting "V6" from the dropdown menu. Alternatively, users can add the parameter "-v 6" after each image description.

Mr. Holz pointed out the following innovations: "much more precise adherence to descriptions and longer descriptions; improved consistency and model knowledge; enhanced prompting and correction mechanisms; some text drawing capabilities—specified in quotes, parameters like '--style raw' and low values of '--stylize' can help; improved scaling tools with 'precise' and 'creative' modes."

The mechanism for creating image descriptions in Midjourney has been completely overhauled: parameters like "photorealism" or "4K" no longer work, and users will need to relearn. Developers advise users to be more explicit about what they want. For example, if a photorealistic image is required, it is recommended to use the "--style raw" parameter. Users should also choose the appropriate balance for the "--stylize" parameter, whose default value is "100." The lower the value, the more literally the model interprets the command, while increasing it enhances the aesthetic aspect of the image.

The head of Midjourney emphasized that V6 is released as part of alpha testing—its functionality may change significantly over time, often without prior notice to the audience. Developers express hope that with the full release of the model, users will "feel the development of something more significant, deeply intertwined with the power of our collective imagination." However, some features present in V5.2 are currently absent in Midjourney V6, including left and right panning and zooming, but they will be added in subsequent updates.