Midjourney V2 AI moderation system is here: more detailed compliance checks for prompts and the editing area

Midjourney has recently been testing a more “big-picture” V2 AI moderation system. It doesn’t just watch the prompt; it also considers the images you upload, the painted masks, and the final generated results as part of its judgment—reducing room for luck and making compliance boundaries clearer.

What exactly has the V2 AI moderation system updated?

In the past, many people got used to splitting risky content across different parts of the prompt or trying to “get around it” via partial re-draws, but Midjourney’s V2 AI moderation system performs an overall check. According to the official description, it evaluates the prompt, the input image, the mask you draw during editing, and the generated output image at the same time, and then decides whether to allow it through or block it.

Because this is early testing, Midjourney has also clearly stated that the rules are still being optimized, so you may encounter cases where similar prompts sometimes pass and sometimes get rejected.

Greater impact on prompts and “partial edits”

If you often use Midjourney’s editing workflow (for example, erasing an area and then filling in content), the V2 AI moderation system will look at “what you erased, what you intended to add, and what ended up being added” together. In other words, the intent within the masked area is easier for the system to understand, and violation risk won’t be overlooked just because you “only changed a small part.”

For normal creation, this is a good thing: compliance judgments become more consistent; for people trying to skirt the edge, Midjourney’s tolerance will be lower.

A more reliable approach: state your intent clearly and lock the boundaries

To reduce moderation volatility in Midjourney, it’s recommended to replace vague wording with more specific, controllable descriptions. For example, change “more sexy” to “fashion photography, tasteful formalwear, business magazine style,” and change “a sense of violence” to “tense atmosphere, movie-poster lighting, no gory details.”

The same applies when editing: Midjourney is more likely to accept clear changes like “replace an object/material/lighting,” rather than repaint requests that point to sensitive subjects or sensitive behaviors.

Usage reminder: during the testing phase, assume it will be “stricter”

At present, Midjourney’s V2 AI moderation system is still in the testing period, and the experience will change as rules are adjusted. If you repeatedly modify the same image, it’s recommended to change only one direction at a time and keep the previous usable version, to avoid being unable to revert when moderation tightens.

In one sentence: Midjourney is upgrading moderation from “looking at one sentence” to “looking at the entire workflow.” The earlier you adapt to this way of writing, the less likely you are to be interrupted at a critical moment when generating final outputs.