The Agency Guide to Midjourney: From Brief to Deliverable

Creative agencies have adopted Midjourney faster than almost any other AI tool. The reasons are obvious: a creative director can generate twenty concept directions in the time it used to take to brief an illustrator on one. Freelancers deliver AI-assisted mood boards and visual explorations that would have been cost-prohibitive two years ago. Studios iterate on brand campaigns at a pace that clients are starting to expect.

But most agencies are still running Midjourney the way they discovered it—ad hoc. Prompts live in personal Discord histories or scattered Notion pages. Iterations are tracked by memory. Client approvals happen over email with no link back to the specific variant that was selected. The generation is fast; everything around it is manual, fragile, and invisible.

This guide lays out a six-phase workflow for taking a Midjourney project from client brief to final delivery package—with the governance, naming conventions, and audit trails that professional work demands.

The Agency Midjourney Reality

The way agencies use Midjourney is fundamentally different from how individual creators use it. An individual generates images for themselves. An agency generates images on behalf of someone else—a client who has brand guidelines, approval workflows, usage rights requirements, and regulatory obligations.

This distinction changes everything about how you need to work. Three common agency scenarios illustrate the point:

• Creative directors use MJ for rapid concepting and mood boards. They need to generate dozens of directions, present a curated shortlist to the client, and trace back to exactly which prompt and parameters produced the approved direction.
• Freelancers deliver AI-assisted visuals as part of larger projects. They need clean handoff packages with usage documentation, and they need to protect their creative process—prompts are intellectual property.
• Studios run multi-campaign, multi-client operations. They need consistent naming, shared asset libraries, and the ability to pick up where a colleague left off without a thirty-minute Slack archaeology expedition.

In every case, the generation itself is the easy part. The hard part is everything around it: organising, selecting, approving, packaging, and delivering with confidence.

The Six-Phase Agency Workflow

Whether you are a solo freelancer or a twenty-person studio, the professional Midjourney workflow has six distinct phases. Most agencies formalise one or two of these and improvise the rest. The gaps are where projects go sideways.

Phase 1: Brief Intake

Every project starts with a creative brief. For Midjourney work, the brief needs to capture more than the usual “look and feel” direction. You need explicit answers to:

• Output specifications — Dimensions, aspect ratios, file formats, resolution requirements
• Brand constraints — Existing style references, colour palettes, visual elements to include or avoid
• Usage rights — Where the images will be used (web, print, social, paid media), geographic scope, duration
• Disclosure requirements — Whether the client needs to disclose AI generation to their audience or regulators
• Iteration budget — How many rounds of revision are included, and what constitutes a “round”

A brief that does not address these points will generate scope creep. You will discover the missing requirements mid-project, when they are expensive to accommodate.

Phase 2: Prompt Development

With the brief defined, prompt development is where creative expertise meets technical knowledge. This phase typically involves:

• Translating the creative brief into initial prompt structures
• Testing style references (--sref) against the brand constraints
• Establishing parameter baselines—aspect ratio, stylize weight, chaos value, model version
• Generating a test grid of 8–12 outputs to validate the creative direction before committing to full production

The outputs from this phase are internal. They exist to prove the direction works before you show anything to the client. Document which prompts and parameters produced which results—you will need this mapping when the client says “more like number three but warmer.”

Phase 3: Iteration Rounds

Client feedback drives iteration. Each round should be structured:

• Present a curated set of options (typically 4–8 per round, not the 40 you generated)
• Label each option with a clear identifier (not “the blue one” but “Direction A, Variant 3”)
• Record the client's feedback against each specific variant
• Track which prompt modifications address which feedback points

The iteration trap is losing track of what the client has already seen and reacted to. Without a system, round three's feedback contradicts round one's, and nobody can reconstruct the decision history.

Phase 4: Client Selection

The client picks their final selects. This sounds simple, but it is where governance gaps cause the most pain. You need a clear record of:

• Which specific variant was approved — Not “the one from Tuesday's email” but a traceable identifier linked to the exact output file
• Who approved it — Name, role, date
• Approved usage scope — The specific channels and contexts where this image is cleared for use

When a client calls six months later asking “which image did we approve for the homepage banner?” your answer should take thirty seconds, not thirty minutes.

Phase 5: Post-Processing

Midjourney outputs rarely go directly to production. The post-processing pipeline typically involves:

• Upscaling — MJ outputs at 1024px; print and large-format digital need higher resolution
• Retouching in Photoshop — Fixing hands, text, brand-specific colour correction, compositing with other elements
• Layout in Figma or InDesign — Integrating the approved image into the final design
• Format conversion — Producing the required file formats and sizes for each delivery channel

Each post-processing step creates a derivative. The governance question is: can you trace from the final delivered file back through the processing chain to the original Midjourney output and the prompt that created it? For commercial work, this traceability is not optional.

Phase 6: Delivery Package

The delivery package is what the client receives. For professional Midjourney work, this needs to be more than a folder of PNGs dropped into Google Drive. A complete delivery package includes:

• Final output files in all required formats and resolutions
• Usage rights documentation clarifying what the client can do with the images
• AI disclosure language the client can use if required by their industry or jurisdiction
• Approval record confirming which images were selected and by whom

What the delivery package should not include: your prompts (unless contractually required), intermediate iterations the client did not select, internal notes and feedback, and any style reference codes that represent your proprietary creative process.

What Breaks Without Governance

Running Midjourney without workflow governance works exactly until it does not. The failure modes are predictable and expensive:

The common thread is traceability. Every governance failure is ultimately a failure to connect a delivered image back to the decisions that produced it.

Naming and Structure Conventions

A naming convention is the cheapest governance tool you have. It costs nothing to implement and prevents the most common retrieval failures. Here is a structure that scales from freelancer to studio:

Project folder structure:

[CLIENT-CODE]/[YEAR]-[BRIEF-NUMBER]/
  01-brief/
  02-prompt-dev/
  03-iterations/
    round-01/
    round-02/
  04-selects/
  05-post-processing/
  06-delivery/

File naming:

[CLIENT]-[BRIEF]-[DIRECTION]-[VARIANT]-[VERSION].[ext]
ACM-2026-03-A-V2-final.png
ACM-2026-03-B-V1-draft.png

This convention encodes four pieces of information into every file name: who it is for, which brief it belongs to, which creative direction it represents, and which variant within that direction. Anyone on the team can look at a file name and understand its context without opening it.

Manual naming conventions work well for solo freelancers and small teams running one or two client projects at a time. They start to break down when you have multiple clients with overlapping campaigns, team members who interpret the convention differently, or project timelines that span months with personnel changes mid-stream.

At that point, the convention needs tooling support—a DAM or project management system that enforces the structure rather than relying on individual discipline.

Compliance for Commercial Use

If your agency delivers AI-generated images for commercial use, compliance is no longer optional. Two developments are reshaping the landscape:

EU AI Act disclosure requirements. From August 2026, AI-generated content used in certain commercial contexts within the EU requires clear labelling. The specifics are still being finalised in implementing acts, but the direction is clear: if you deliver AI-generated visuals to clients operating in the EU, you need to be able to demonstrate what was generated by AI and provide appropriate disclosure language.

IPTC Digital Source Type. Midjourney embeds trainedAlgorithmicMedia as the IPTC Digital Source Type in exported images. This is a start—it marks the image as AI-generated at the metadata level. But IPTC Digital Source Type alone does not constitute full compliance. It does not record who generated it, when, for what purpose, or under what terms. Those records need to come from your workflow.

Practical steps for agencies:

• Maintain generation records — Date, prompt, parameters, model version, and the person who ran the generation
• Include disclosure language in delivery packages — A simple statement that the images were created using AI image generation tools, with the specific tool identified
• Preserve metadata through post-processing — Photoshop and other tools can strip IPTC data during editing; ensure your post-processing pipeline preserves or re-embeds it
• Discuss disclosure with clients proactively — Do not wait for the client to ask; include AI disclosure as a standard item in your project kick-off

Scaling from Freelancer to Studio

The workflow described above works for a solo freelancer with discipline and a good folder structure. But each step up in team size introduces new coordination problems:

• Shared prompt libraries — When two designers work on the same client, they need access to the same prompt foundations. A personal prompt library becomes a shared resource that needs versioning and access control.
• Approval workflows — A freelancer approves their own work. A studio needs a creative director to review outputs before they go to the client. This review step needs to be part of the workflow, not an afterthought.
• Brand consistency enforcement — With multiple people generating images for the same brand, style drift is inevitable without explicit governance. Approved style references and parameter presets help, but they need to be maintained and distributed.
• Client isolation — One client's assets, prompts, and style references should never be visible to another client's project team. This is trivial for a freelancer with one client and non-trivial for a studio with twenty.
• Onboarding — When a new designer joins the team, how do they discover approved styles, understand naming conventions, and access the prompt libraries for their assigned clients? If the answer is “ask Sarah,” your workflow does not scale.

The inflection point typically arrives around three to five team members working on concurrent client projects. Below that, manual conventions and good communication carry you. Above it, you need systems that enforce the conventions automatically—a DAM with tagging, access control, and search; a project management tool integrated with your asset pipeline; or both.

Numonic addresses several of these scaling challenges—tenant isolation, visual search, metadata preservation, and governed sharing. But regardless of which tools you choose, the workflow structure remains the same. The six phases do not change. What changes is whether humans enforce them or software does.

The Workflow Is the Product

Agencies that treat Midjourney as “just another tool” will generate images quickly and lose time everywhere else: finding approved variants, reconstructing decision histories, assembling delivery packages, answering compliance questions. Agencies that treat the workflow as a first-class concern will deliver faster, with fewer errors, and with the audit trails that professional clients increasingly require.

The six-phase workflow is not overhead. It is the difference between running a professional operation and running a creative hobby with invoices attached. The generation step—the part where Midjourney does its work—is maybe twenty percent of the total effort. The other eighty percent is brief capture, iteration management, approval tracking, post-processing, and delivery packaging. That eighty percent is where your professional value lives.

Start with the phase you are weakest on. For most agencies, it is either iteration tracking (Phase 3) or delivery packaging (Phase 6). Fix that one phase, and the improvement in client confidence and team efficiency will justify investing in the rest.