Midjourney V7

What it is

Midjourney V7 is an image generation model from Midjourney, released to the company's community as an alpha on 4 April 2025. Midjourney described it as "an entirely new model" rather than a refinement of what came before ^[1]. It followed V6, and at launch it was offered in Turbo and Relax modes only; a Turbo job cost twice what a normal V6 job cost ^[1].

Releasing a model as an alpha means it goes out before everything is finished, and V7 matched that pattern: some of the tooling around the model still ran on the previous generation when it opened.

Personalization, on by default

V7 was the first Midjourney model with personalization switched on by default ^[1]. A user unlocks it by rating images — about five minutes of rating is enough — and the feature can also be turned off ^[1]. The point of the ratings is to give the model a picture of what that particular user likes, so that the same prompt can be steered toward different results for different people. Midjourney said personalization raised the bar for how well the model could interpret what a user wants and what they find beautiful ^[3].

Draft Mode and a conversational prompt bar

The biggest change to the working method was Draft Mode, which renders at half the cost and ten times the speed of a standard job ^[1]. A user could trigger it explicitly by putting `--draft` after the prompt, and draft images came out at lower quality than standard jobs ^[3]. The trade-off is deliberate: cheap, fast pictures for trying out an idea, with the option of committing to a full-quality render once something looks right.

On the web, draft mode also came with a conversational prompt bar and a voice mode ^[1]. The prompt bar let users ask for changes in ordinary language — swapping a cat for an owl, for example — and clicking "enhance" or "vary" re-rendered a chosen image at full quality ^[3]. That turns prompting from a single command typed once into a back-and-forth in which the user reacts to what came back.

Quality claims and gaps

Midjourney said V7 was much smarter with text prompts and more coherent for bodies, hands and objects ^[1]. Text prompts matter because they are the main channel through which a user tells the model what to draw, so better prompt handling means fewer rounds of rewriting before a usable image appears. Not everything shipped on day one, though: upscaling, editing and retexture initially fell back to V6 models, and character and object references were planned but not yet available ^[1]. In practice that meant users could generate with the new model but had to drop back to the old one to finish some jobs.

Where it sits

V7 was one stage in a fast-moving line of releases: its successor, V8, reached alpha in March 2026 ^[2]. V7 is the version where personalization became the default setting, where a low-cost draft tier appeared alongside the full-quality one, and where the web interface began to accept follow-up instructions in plain language and by voice.