Midjourney V1

What it was

Midjourney V1 was the first image-generation model released by Midjourney, in February 2022 ^[1]. It was a closed model: rather than being something users could download and run on their own machines, it was available only through Midjourney's own service ^[1]. That service was reached through a bot on Midjourney's Discord server, where a user sent a request in a chat channel and the bot returned an image; the public open beta on that bot began later, on 12 July 2022 ^[1].

Midjourney's own documentation records that Version 1 was the default model from February 2022 to April 2022 ^[2]. In other words, for the first few months of the service, anyone who used it without asking for a particular version was using V1.

What its images were like

Midjourney describes Version 1 as very abstract and painterly, with low coherency ^[2]. Coherency here refers to whether the parts of a picture hold together — whether the light comes from one direction, whether a figure has the right number of limbs, whether an object stays the same object across the frame. A model with low coherency tends to produce images that read as suggestions of a subject rather than as careful renderings of one. The two descriptions go together: a painterly, abstract look is what you get when the model is strong on texture and mood and weak on the precise arrangement of things.

Resolution was modest by later standards. Midjourney's upscaler table groups Versions 1 to 3 in the same row: a 256 x 256 pixel starting grid, a 1024 x 1024 pixel default upscale, and a maximum upscale of 1664 x 1664 pixels ^[2]. The starting grid is the small image the model actually generates; the upscaler is a second step that enlarges it. So a V1 image began as a 256-pixel-square picture and was then blown up, four times over by default, with the upscaler filling in detail.

Naming

One small piece of history survives in Midjourney's list of deprecated parameters, which notes that `--vibe` is now known as model version 1 ^[2]. The parameter was a text tag typed alongside a request to select a model; the tag that once meant "the current model" was later renamed to name the first version specifically. That is a common pattern in fast-moving tools: as new versions arrive, whatever used to be the default has to be given a number.

Where it sits in the line

V1 did not stay current for long. Midjourney released V2 on 12 April 2022 and V3 on 25 July 2022 ^[1]. The handover is visible in the documentation as well: V1's spell as the default model ran from February to April 2022, ending at the point when V2 arrived ^[1][2].

V1 matters mainly as a starting point. It is the version that established what the product was — a hosted, closed model reached by typing into a chat bot rather than a piece of software you install — and it is the baseline against which the following versions were measured, including in Midjourney's own tables, where it is still listed alongside V2 and V3 with the same resolution figures ^[2]. Its weaknesses are part of that record too: low coherency and a heavily stylised, abstract look are the problems that later versions were positioned against.