AI&Design
All change articles

Midjourney · Announcement · Official source

What changed when Midjourney made V7 the default

On June 17, 2025 at 04:46:33 UTC (13:46:33 in Seoul), Midjourney announced that users who had yet to try V7 would be switched to it automatically. V6.1 remained selectable in web or Discord settings or for an individual prompt with --v 6.1. Midjourney described V7 as producing higher-quality images, following prompts more closely and rendering more coherent details. The archived announcement gives those claims without comparative scores. June default-model announcement · Midjourney archive

Midjourney had already announced the V7 alpha on April 4 at 04:19:23 UTC (13:19:23 in Seoul) and described similar gains in text and image prompting, texture, and the coherence of bodies, hands and objects. The current Version article lists April 3 as the release date and June 17 as the start of the default period. The April 4 UTC announcement timestamp still falls on April 3 in US Pacific time. A separate model update on April 30 at 22:59:14 UTC (May 1 at 07:59:14 in Seoul) claimed further gains in image quality, prompt accuracy, hands and bodies. June 17 is the date those already available V7 capabilities moved onto the default generation path. Alpha announcement · April model update · Midjourney archive · Version documentation

What observers measured and saw

On June 11, six days before the default switch, HobbesSR reported a frequency analysis of V6, V6.1 and V7. The published workflow used 50 nonsense prompts and four 1024 × 1024 outputs per prompt from each model, for 600 images. It converted images to grayscale without resizing, computed a two-dimensional Fourier transform, and summarized normalized spectral magnitude by radial frequency. The author reported mean shares of 1.65%, 1.50% and 1.46%, respectively, in the highest fifth of the frequency range. Fine detail and noise both contribute to that statistic. The author acknowledged that a reduction in noise could also explain the drop. Random text gives this comparison no way to score whether an image followed a meaningful brief or had coherent anatomy. June 11 original post · Dataset and analysis repository

John Angelo Yap repeated text prompts in V6.1 and V7 and showed a pair of outputs for each of seven categories. He found more convincing motion and photographic detail in the V7 marathon runner and more natural-looking waves in the V7 beach Polaroid. In the water bottle scene he found the V7 lighting less effective, and for the comic villain he found V6.1 closer to the requested aesthetic. Both displayed street signs garbled the requested wording. The article preserves prompt text and examples, while seed, personalization, sampling and selection details are unspecified. The site marks the article updated June 13, 2026. Its original publication and render dates are unknown. Gold Penguin’s paired prompt examples

On July 27, a Reddit user showed a V6.1 and V7 pair from the same text prompt and described the V7 image as more cartoon-like, with distracting foreground detail. A respondent reran the prompt in V7 with their own personalization profile, stylization at 150, experimental aesthetics at 20, Raw enabled and quality at 4. The rerun documents a practical way to tune an output. Several controls changed together, so the replies leave each setting’s effect undetermined. July 27 paired comparison and settings

What the launch-period reports establish

On June 30, a returning user reported repetitive images and poorer perceived prompt fit despite trying personalization on and off and switching to V6.1 or V6. A regular V7 user in the same thread reported diverse results, while others raised concerns about textures and anatomy. The original post supplies no paired jobs, and its report spans multiple versions. The June evidence establishes when the default changed, a measured texture-related difference in one pre-switch dataset, and disagreement among users about the outputs they were getting. The available comparisons leave the size and consistency of any improvement in prompt adherence or image coherence unresolved. June 30 discussion · June 11 frequency comparison · June default-model announcement