Midjourney has launched V8 Alpha on alpha.midjourney.com, marking the biggest upgrade to its image generation platform since V6.
The headline feature is native HD image generation — users can now produce 2K resolution images using the --hd parameter without any upscaling step. This eliminates the quality loss that previously came from post-processing enlargement.
Speed improvements are equally dramatic. V8 Alpha renders standard jobs approximately 4-5x faster than earlier versions, making it the fastest Midjourney model to date. The Alpha website UI has been upgraded to match, with settings, image references, Personalization profiles, moodboards, and a new grid view accessible to the right of the Imagine bar.
Text rendering sees a major leap forward. When users place text in quotation marks within prompts, V8 renders it with significantly improved accuracy — readable street signs, clean product labels, and legible typography in posters and book covers are now achievable.
Prompt following is substantially improved overall. Complex multi-element compositions render with noticeably higher fidelity to the original prompt. Users can toggle Raw mode for even more prompt adherence by removing default styling.
A new --q 4 quality mode provides additional image coherence for complex scenes, helping maintain visual consistency across the entire image.
However, early user testing has revealed mixed results. While landscapes, architecture, and product shots show clear improvements, generating accurate hands and very specific complex details remains inconsistent. The community has noted that V8 excels at broad creative direction but still struggles with anatomical precision in certain scenarios.
Midjourney also confirmed the switch from TPUs to GPUs and PyTorch for V8's infrastructure — a move that opens up hiring and development since more engineers are familiar with the PyTorch ecosystem.