The Midjourney Delivery Package: Hand Off AI Images Without Losing Context—or Oversharing It

You generated the images. The client approved the concepts. Now you need to deliver. For agencies and freelancers using Midjourney, this is where things get awkward. Every PNG you downloaded from midjourney.com contains your full prompt, your username, generation parameters, and an AI-generated disclosure tag—all embedded in the file metadata. Send those files as-is and you are handing over your creative process alongside the creative output.

Most delivery problems are not about the images. They are about what travels with the images. This guide establishes a standardised delivery workflow that gives clients everything they need—high-resolution finals, usage context, AI disclosure—without exposing what they should not see.

What Clients Actually Need

Start with the client's perspective. When receiving AI-generated images, clients need four things:

• High-resolution finals — The approved versions at full resolution, in the agreed format (PNG, TIFF, or converted to JPEG for web use). Not every variation. Not the exploration grid. Just the approved outputs.
• Usage context — Where and how the images can be used. This is especially important for AI-generated content, where usage rights differ from traditional photography or illustration.
• AI disclosure — Increasingly, clients need to know (and disclose) that images were AI-generated. The EU AI Act enforcement begins August 2026. Some industries and platforms already require AI labelling.
• Technical specifications — Resolution, colour profile, aspect ratio, file size. Standard deliverable information that the client's design team needs for production.

What Clients Should Not See

Just as important as what you include is what you exclude. These are not trade secrets for the sake of secrecy—they are professional boundaries that protect both parties.

• Your prompts — These are your creative methodology. A prompt like “editorial fashion photography, Kodak Portra 400, golden hour backlight, magazine cover composition, --ar 3:4 --stylize 800 --v 7” reveals the technical craft behind the output. Clients pay for results, not recipes.
• Your Midjourney username — The Author field identifies you personally. Depending on the relationship, this may or may not be appropriate in delivered files.
• Intermediate iterations — The fifty variations you generated before finding the right direction are part of your process, not the deliverable. Sharing them invites scope creep and second-guessing.
• Internal notes and project context — Any annotations, file naming conventions, or organisational structure that reveals your workflow or references other clients.

The Delivery Package Checklist

A professional Midjourney delivery package contains five elements. Anything less raises questions. Anything more risks oversharing.

1. Finals folder — Approved versions only. Named clearly (not Midjourney's default UUID filenames). Metadata stripped of prompt and author fields.
2. Usage rights document — A plain-text or PDF statement covering: licence scope (exclusive, non-exclusive, perpetual, limited), permitted use cases (web, print, social, advertising), and any restrictions. Midjourney's current terms grant commercial usage rights to paid subscribers, but your client contract should specify what you are transferring.
3. AI disclosure statement — A brief document confirming the images were generated using AI (specifically Midjourney), noting the IPTC Digital Source Type tag (trainedAlgorithmicMedia) is preserved in delivered files for machine-readable AI identification.
4. Technical specifications — Resolution (pixels), file format, colour profile, aspect ratio, and file size for each deliverable. This can be a simple spreadsheet or a section in the delivery document.
5. Revision history (sanitised) — A brief log of approved rounds: “R1: Initial concepts (3 directions). R2: Refinement of Direction B. R3: Final with colour adjustment.” No prompt details, no variation counts, no technical parameters.

Metadata Management for Delivery

This is the most important technical step. Every Midjourney download contains identical metadata whether downloaded individually or in a batch ZIP. You need to selectively strip fields before delivery.

The ExifTool command for delivery-ready files:

# Strip prompt, author, and GUID; keep IPTC source type and creation time
exiftool -Description= -Author= -DigitalImageGUID= \
  -overwrite_original /path/to/delivery/finals/

Run this on the delivery folder after copying files from your master archive. Never modify your originals. See our privacy sharing guide for the full framework on when to strip and when to preserve.

Template: Client Delivery Folder Structure

Consistency signals professionalism. Use the same structure for every delivery so clients know what to expect and your team knows what to prepare.

/client-name/
  /campaign-name/
    /finals/
      hero-banner-3840x2160.png
      social-square-1080x1080.png
      product-detail-2400x3200.png
    /disclosure/
      ai-disclosure-statement.pdf
      usage-rights.pdf
    /specs/
      technical-specifications.csv
      revision-history.txt

Naming conventions for finals: Replace Midjourney's default UUID filenames with descriptive names that match the client's brief. A file named hero-banner-3840x2160.png communicates purpose and specs at a glance. A file named dr.jb_a_cinematic_landscape_with_dramatic_clouds_abc123.png reveals your username and prompt in the filename itself—a privacy leak that metadata stripping alone will not fix.

Version naming: If delivering multiple rounds, use v1, v2, v3 suffixes. Never use Midjourney variation numbers (V1–V4) or upscale labels (U1–U4) in client-facing filenames. These expose your generation workflow.

Handling “Can I See the Prompt?”

Clients sometimes ask. The question is reasonable—they want to understand how the work was made, or they want to generate similar images themselves. How you respond depends on the relationship and the contract.

Option 1: Polite decline. “Our prompts are part of our proprietary creative methodology. We are happy to generate additional variations if you need a different direction.” This is appropriate when your prompt engineering is a core part of the value you deliver.

Option 2: Selective sharing. Share a simplified version of the prompt that conveys the creative direction without revealing your specific techniques. Instead of the full prompt with parameters, style references, and weighting, share: “Cinematic landscape photography with dramatic lighting.” This gives the client context without giving away your process.

Option 3: Full disclosure under NDA. For ongoing retainer clients or situations where prompt sharing is contractually required, share the full prompt under a non-disclosure agreement. Document what was shared, when, and under what terms.

Your Internal Record: The Other Half

The delivery package is what the client sees. But your internal record must be richer. For every delivery, your archive should contain:

• Master files with full metadata — Every image as downloaded from Midjourney, with prompt, author, Job ID, and all parameters intact. This is your searchable creative record.
• All iterations, not just finals — The variations, upscales, and rejected directions that led to the approved output. These are invaluable for future projects with the same client or similar briefs.
• Client brief and approval chain — What was requested, what was approved, and when. This protects you if there are disputes about deliverables.
• Delivery log — A record of what was sent, when, to whom, and with what metadata configuration. If a client later asks “did you send me the high-res version?” you have the answer.

This dual-track approach—stripped copies for delivery, full-metadata originals for your archive—is the core pattern. A post-export management system makes this sustainable at scale by maintaining the master archive and producing delivery-ready exports on demand, with configurable privacy presets per client or use case.