Midjourney v8.2

Midjourney v8.2 is the one Mixen image model that is not callable over /v1. It lives in two interfaces:

• web cabinet — mixen.ai/account/images, the Midjourney section;
• Telegram bot — the image mode with Midjourney selected.

It is absent from the /v1/models catalog on purpose: the midjourney-v7-* keys are billing SKUs (mode × action), not API models. Sending such a key to /v1/images/generations returns 404 model_not_found.

How it works

Midjourney always answers with a grid of four options. Each tile then supports post-actions:

Stage	What happens
Imagine	A 2×2 grid generated from the prompt in native HD (--hd): the finished grid is 2048×2048, each tile 1024×1024
Upscale (U1–U4)	The chosen tile at full 2048×2048 in one step — v8 has no separate Subtle/Creative upscalers, HD is rendered natively
Variation (V1–V4)	Four new variations of the chosen tile
Reroll	A full re-run of the prompt — a fresh grid

Besides text-to-image there are two modes with images on input: Blend mixes two to five uploaded images (cabinet and bot), and Describe writes a prompt that reproduces an uploaded image (bot).

Parameters

Parameter	Values	Where
Generation mode	relax — cheap, queued; fast — price/speed balance (default); turbo — fastest	cabinet and bot
Frame format	--ar in the prompt: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, 9:21, 5:4, 4:5	bot and cabinet
Blend format	PORTRAIT, SQUARE, LANDSCAPE	cabinet
Post-actions	upscale U1–U4, variation V1–V4, reroll	cabinet and bot

The format is set with the usual Midjourney flag at the end of the prompt. If the flag is already written there, the format chosen in the interface does not override it — your text wins. Version and resolution are appended automatically as well (--v 8.2 --hd); your own --v, --niji, --hd/--sd in the prompt take priority. In the bot the blend format is not picked separately: it is derived as the closest of the three from the current frame format.

Blending and reverse prompts

Blend takes two to five images and mixes them into one scene — works in both the cabinet and the bot. Typical use: combine a character from one image with a background from another.

Describe (bot) solves the inverse problem: send a finished image and get a prompt that reproduces it. Handy for capturing a reference’s style and then iterating on it in text.

What lives where

Feature	Cabinet	Bot
Imagine (2×2 grid)	yes	yes
Upscale / Variation / Reroll	yes	yes
Relax / Fast / Turbo speed choice	yes	yes
--ar frame format	yes	yes
Blend (2–5 images)	yes	yes
Explicit blend format choice	yes	no (derived from the frame format)
Describe (reverse prompt)	no	yes

Prompt examples

cyberpunk street market at night, neon reflections, rain --ar 21:9
minimalist poster for a coffee shop, warm tones --ar 4:5

A typical pipeline: generate a grid → pick the good tile → U to enlarge or V for variations → reroll the whole prompt if needed. In the bot, cutting a tile U out of the grid happens locally and is free — only fetching the tile at 2048×2048 is billed.

Price

Midjourney prices are not published in /v1/models — the model is not in the catalog. The grid and the post-actions are billed separately; the bot shows current amounts in the legend under the finished grid.

Strengths and limits

• The recognizable Midjourney artistic style — the reason it gets picked at all.
• A 2×2 grid per generation: four options to choose from without repeating the request.
• U/V/Reroll post-actions on a finished grid — precise refinement without a new prompt.
• Blend with up to five images, and Describe for reverse-engineering a prompt from an image.
• No /v1 access — it cannot be automated in a pipeline, only through Mixen interfaces.
• No catalog prices; the bot is the fastest place to see them.

All the other models — in the catalog. Image generation over the API — images guide.