Midjourney v8: An Ambitious Comeback in Image Generators

Midjourney v8: A Strategic Return to the Forefront

Since spring 2025, Midjourney has seen its models drop out of the top ten image generators according to the LMArena ranking. In response to this situation, the startup led by David Holz decided to take action. On March 17, 2026, it announced the alpha release of its v8 model, touted as a complete overhaul, accessible via a dedicated website. This launch comes nearly a year after a version 7 that failed to impress, and just a few months after the introduction of an AI-powered video generator.

Enhanced Performance and High-Resolution Images

The alpha version of Midjourney v8 stands out for its speed, being described as the fastest model ever offered by the company. According to the publisher, the image generation speed is five times faster than previous versions. To accompany this improvement, the web interface has been redesigned, featuring an enhanced chat mode, a grid view, and sidebars for accessing settings without interrupting the creative process.

In terms of image quality, Midjourney highlights several significant advancements. The v8 model better understands detailed prompts and provides more reliable text rendering, provided that the text is enclosed in quotes within the prompt. The produced images are also more coherent and detailed, with increased customization options through profiles, style references (srefs), and moodboards.

A notable addition is the –hd mode, which allows for native generation of 2K resolution images without the need for upscaling. Additionally, the –q 4 mode is introduced to improve image consistency. Customization profiles, moodboards, and srefs created with version 7 remain compatible with this new version.

Limited Availability and Cost of Advanced Features

Currently, v8 is accessible only via the site alpha.midjourney.com. It is not yet available on the main Midjourney site or on Discord, and creations made with this model are not visible on the usual platform.

Regarding pricing, the most anticipated features come at a high cost: generations using the –hd mode, –q 4, style references, or moodboards consume four times more GPU time than a standard generation. The Relax mode, which allows for image generation without extra cost but at a reduced speed, is not available at launch. Midjourney is working on a new server cluster to offer this option later, along with less costly rendering modes.

A Model Still in Development Facing Increased Competition

The advanced modes –hd, –q 4, style references, and moodboards incur a cost four times higher than a standard generation. The combination of the –hd and –q 4 modes in a single generation can multiply this cost by sixteen.

This launch comes as Midjourney continues to rely on a diffusion-based model, while several competitors have integrated autoregressive components into their image generation pipelines, giving them an edge in prompt accuracy. The startup presents this v8 as a completely new model, with unique strengths and weaknesses, and encourages its community to test and evaluate the generated images to accelerate the model's improvement. No specific timeline has been provided for the release of the final version.