
Midjourney has announced a new image model, V7, that is rolling out in alpha starting today. This new model, according to Midjourney, brings better image quality with "beautiful textures," better coherence for things like hands and objects, among others. The model ships with two modes, Turbo and Relax, with Turbo costing more than twice as much as regular V6 mode.
There's also a new feature called "Draft mode," which costs half as much as standard mode while rendering images ten times faster.
There are two ways to use Draft mode. The first involves clicking "draft mode," then "voice mode" and speaking your ideas. The second involves using the --draft flag right after your prompt.
Images generated with Draft mode are lower quality compared to standard mode. If you see something you like in Draft mode, you can click "enhance" or "vary" on the image to render it in full quality.
Tasks like mood boards and Style Reference Codes (SREF) work with V7, while for tasks like upscaling, editing, and retexturing, V6 models will be used. Availability in V7 models is slated for the future.
If you're interested in V7, keep in mind that you must unlock your personalization to use it since the model ships with personalization turned on by default. In Midjourney, personalization means the AI learns your style and preferences over time. Instead of just generating random images based on your prompts, it tries to match what you like.
To make this work, you have to rank at least 200 pairs of images. The whole process takes five minutes. This helps the AI understand your taste. Once that’s done, you can turn personalization on or off anytime.
Midjourney has come a long way since it was founded by David Holz back in 2022. In its early days, the platform ran entirely on Discord, meaning you'd have to be part of the Midjourney Discord server to generate images with its models. If Discord's not your thing, the official website is available, though you'd need a subscription.
Midjourney faces tough competition from AI labs like OpenAI, which recently launched a viral new image generation tool that is very good at rendering text, a common weakness for image generation models. OpenAI's new tool is so good that some users have discovered you can use it to generate fake receipts.
1 Comment - Add comment