AI image generation spans a rich ecosystem of tools — Midjourney for artistic output, Stable Diffusion (via AUTOMATIC1111 or ComfyUI) for local, fully-configurable pipelines, and OpenAI's GPT-image models for instruction-following and text-in-image work. Understanding prompt anatomy, model parameters, and conditioning tools like ControlNet separates predictable, professional results from random rolls of the dice. The single most important mental model: every generator is a two-sided steering wheel — the positive prompt pushes toward a target, while negative prompts, parameter settings, and reference images all push away from unwanted territory; learning to drive both sides simultaneously is the skill.
What This Cheat Sheet Covers
This topic spans 15 focused tables and 137 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
Table 1: Midjourney Core Parameters
Midjourney parameters are appended at the end of a prompt and directly control how the model renders, stylises, and varies images. These are the most-used parameters — knowing their defaults and ranges lets you move from "hope it works" to intentional steering.
| Parameter | Example | Description |
|---|---|---|
--ar 16:9--ar 3:2 | • Sets width-to-height proportions • default is 1:1 (square). Common values: 1:1, 3:2, 4:3, 9:16, 16:9. | |
--s 100--s 750 | • Controls artistic interpretation intensity; range 0–1000, default 100. • Low (0–50): literal, minimal flair • High (500–1000): strong aesthetic choices that may drift from the prompt | |
--v 7--niji 7 | • Selects model version • V7 is the default as of Q3 2025. Cannot combine --v and --niji in the same prompt | |
--c 30--c 80 | • Controls diversity between the 4 grid images • range 0–100, default 0. Above ~50, the model starts ignoring prompt chunks | |
--no trees--no text | Negative prompt for Midjourney — tells the model to exclude specific elements from the image. | |
--q 1--q .5 | • Rendering quality • values 0.25 (4× faster), 0.5 (2× faster), 1 (default). V6 removed --q 2. | |
--seed 12345 | • Pins the initial random noise to reproduce similar compositions • range 0–4294967295. Cross-session reproducibility is not guaranteed as model updates change seed behaviour | |
--iw 1.5--iw 0.5 | • Balances an image URL prompt against the text prompt • range 0–3 (V7+), default 1. Higher values make the output more similar to the reference image |