Midjourney v5 in the Concept Room: Our Rules of Engagement

Midjourney v5 shipped a few weeks ago, and the difference is not subtle. The uncanny hands are mostly fixed, the lighting is coherent, and the output has crossed the line from "interesting texture" to "could plausibly appear in a pitch deck." Which is exactly why our brand team needed to stop experimenting quietly and write down some rules. This post is those rules, along with the honest observations behind them.

Where it genuinely helps

Let me start with credit where due, because the reflexive dismissal from parts of our industry is not honest either.

Mood and territory exploration is dramatically faster. The first week of an identity project used to involve hours of image research, hunting for photography that approximates a feeling you can see in your head but cannot yet point to. Generating disposable reference images that chase that feeling directly has compressed days into an afternoon. The images are scaffolding; they exist so the team and the client can agree on a direction before anyone commissions the real thing.

It is a brutal cliché detector. Type your concept into a prompt. If the machine, which has ingested the visual internet, produces exactly what you had in your head, your idea is an average. That is genuinely useful information at the concept stage, delivered faster and cheaper than a competitive audit. We now half-joke that a direction is not interesting until Midjourney fails to produce it.

Internal comps for impossible photography. Mocking a hypothetical campaign world to test a territory internally, before committing photo budget, is a legitimate accelerant.

The lines we drew

Now the rules, which exist because capability is not permission:

• Nothing generated ships as final creative. Not in identity work, not in campaign imagery, not "just for the case study." Final work is made by people we can name.
• Clients are told when generated images appear in process work. A mood board slide with generated content says so. Trust is not a place to arbitrage.
• No prompting in living artists' names. Typing a working illustrator's name into a prompt to approximate their style, then not hiring them, is not a gray area for us. It is the specific thing wrong with this moment, and we will not do it. We reference artists the honest way: by commissioning them.
• Nothing client-confidential goes into a prompt. Unreleased names, products, and strategy stay out of third-party systems. Same policy our engineers adopted, same reason.

The unresolved parts, stated plainly

We are not going to pretend the ethics are settled. These models were trained on the work of photographers and illustrators who were not asked, and the legal questions are open: the US Copyright Office signaled just weeks ago that purely generated images do not get copyright protection, which alone should give any brand pause about shipping them. There is a real scenario where the commercial photography market's middle tier gets hollowed out while the top tier and the prompt-typers survive. We do not know how that plays out, and anyone who claims to is selling a webinar.

What we can control is our own practice: use the tools where they compress busywork, disclose their presence, keep money flowing to the human artists whose vision is the actual product, and revisit these rules quarterly, because at the current pace, quarterly may not be often enough.

The takeaway for brand teams

The interesting question is not "will AI make our brand imagery?" It is "what does our imagery mean when anyone can generate plausible pictures?" Our answer: provenance becomes part of the brand. Real photography of real products, real customers, real craft, verifiably made by someone, is about to become a differentiator precisely because plausible is becoming free. The brands that lean into the authentically made thing will stand out in a sea of frictionless, weightless, generated gloss. Plan for that world. It is arriving faster than the hands got fixed.