How to document Midjourney work for client delivery

Midjourney turned prompt-driven image generation into a production staple, and it quietly runs one of the better generation archives in the business. Every job you run is kept on midjourney.com, where the Organize page acts as the central hub for your creations and lets you search by prompt text and by specific attributes such as parameters and individual reference images. For a producer trying to answer the question of how an image was made three months ago, that archive is genuinely useful, as long as the account that made the image is still yours to open.

The catch is that none of this is a delivery record. What your client receives is an image file, and the prompt, the parameter flags, the seed, the model version, and any reference images stay behind in the account. Midjourney also has two properties that surprise teams doing confidential work. First, creations are publicly discoverable by default: Midjourney states that every image and video you create can potentially be found on the Explore page, including work generated in private Discord servers and in direct messages with the bot, and Stealth Mode is limited to the Pro and Mega plans. Second, removal is not what it sounds like. Trashing a creation hides it from your own pages but does not delete it, and Midjourney documents that individual creations cannot be deleted at all short of a full account data deletion.

Seeds deserve one more caution before you rely on them. Midjourney lets you lock a seed and retrieve the seed of a past image, but it also warns that seeds should not be relied on for the same results across different prompting sessions. A saved seed is evidence of how a job was configured, not a guarantee that the image can be regenerated. The checklist below treats the Midjourney archive as a working tool and builds the durable record outside it.

What Midjourney stores on its side

• Midjourney uses a new random seed for every job unless you lock one, and a locked seed is a whole number between 0 and 4294967295 set in the prompt. Source
• You can retrieve the seed of an existing image on midjourney.com through the options button near the prompt text, using Copy and then Seed. Source
• The Organize page on midjourney.com is the central hub for your creations, with search that covers prompt text as well as attributes such as parameters and individual reference images. Source
• Trashing a creation hides it from your Create and Organize pages but does not delete it, and Midjourney states that you cannot delete individual creations short of a full account data deletion. Source
• Every image and video you create can potentially be found on the public Explore page, including work generated in private Discord servers or in direct messages with the Midjourney Bot. Source
• Stealth Mode, which controls who can see your images and videos on the Midjourney website, is available only to Pro and Mega Plan members. Source

What gets lost without your own record

• The delivered file is pixels only. The prompt, the parameter flags, the seed, the model version, and the links to any reference images all stay inside your Midjourney account rather than traveling with the download.
• Midjourney itself warns that seeds should not be relied on to reproduce the same results over different prompting sessions, so even a perfectly kept seed does not promise an identical regeneration later.
• The archive is only as durable as the subscription and the login. If a freelancer generated the work under their own account, the searchable history sits in an account your studio does not control.
• Image references, style references, and character references are searchable inside Organize, but nothing in the delivered file discloses that reference material shaped the result.
• There is no field anywhere in Midjourney for the brief, the client, the rejected takes, or the approval, so the context a dispute would turn on never exists inside the tool.

The documentation checklist for Midjourney

1. Copy the complete prompt including every parameter flag, because settings like --v, --ar, --stylize, and --chaos change the output as much as the words do.
2. Copy the seed right after the job finishes, using the options button near the prompt text on the web and then Copy and Seed, and store it next to the prompt.
3. Record the model version in use at generation time, since the same prompt and seed behave differently between versions.
4. Save the actual files or URLs of every image reference, style reference, and character reference the prompt pointed at.
5. Note the action chain from the first grid to the delivered file, including which variation you chose and which upscale produced the final.
6. Record the visibility mode of the account at generation time, because on plans without Stealth Mode the work is publicly discoverable and a confidential client needs to know that.
7. Write down the generating account, the date, and the link to the job on midjourney.com, so the archive entry can be found again while the account exists.

For EU deliveries, Midjourney work falls under the transparency obligations of Article 50 of the EU AI Act like any other generated imagery, and the penalty ceiling for transparency violations is up to €15 million or 3% of worldwide turnover. Midjourney marks nothing for you and exports no disclosure record, so the statement of what was AI generated, where it was disclosed, and when has to come entirely from your own documentation.

Treat the record you build as a notary rather than a witness. It can show which prompt, seed, references, and outputs existed on which date and that they have not changed since. It cannot show that a specific reference caused a specific pixel, and no honest record claims that. Kept that way, the Midjourney archive stays a convenience for daily work, while the delivery record stands on its own for as long as the client relationship needs it.

Frequently asked questions