The Agency Image Workflow: Producing Visuals for Dozens of Clients at Once

If you run images for one website, image production is a task. If you run them for thirty client accounts, it becomes an operations problem — and the failure mode is not usually quality. It is drift: files that end up in the wrong client folder, a photographer invoice nobody budgeted for, three sites that look like the same stock library, and a metadata field still carrying the agency's name eighteen months after the client left.

The short answer for what this costs, as of August 2026: platform tooling that generates images with SEO metadata included runs roughly $19 to $99 a month depending on volume, which across several client accounts is comfortably under a dollar per finished image. The expensive part is everything around generation. This is a piece about that everything.

The three decisions that determine everything downstream

Before any prompt is written, an agency has to answer three questions, and answering them badly is what produces the mess later.

Where do assets live? Either in the client's own storage or in yours. If they live in yours, offboarding becomes a project. If they live in the client's, your team needs access to a dozen different Drives and DAMs. Most agencies land on a hybrid: production happens in agency space, delivery copies land in client space, and the delivery copy is the canonical one.

Who signs off? Image approval is where timelines die. A client stakeholder who reviews sixty images one at a time will take three weeks. A client who approves a visual direction once and then spot-checks output takes an afternoon. Structure the approval at the direction level, not the file level.

Whose name is in the file? Every generated image can carry a business name, copyright string, contact details and location coordinates in its metadata. Those fields should describe the client, not you. This sounds cosmetic and is not — see the ownership section below.

Structuring accounts so clients cannot contaminate each other

The single most common technical error in multi-client image work is metadata bleed: settings configured for one account persisting into the next batch. It happens because the operator is fast and the defaults are sticky.

Guard against it structurally rather than by attention. A workable pattern:

Layer	What it holds	Who changes it
Agency account	Billing, seat management, global export defaults	Ops lead only
Client sub-account	Business name, copyright string, contact URL, geo-coordinates	Account manager, at onboarding
Brand preset	Palette, setting, camera distance, subject rules	Creative lead, reviewed quarterly
Batch	The specific prompts for one page set	Producer, per job

The point of the table is that the fields most likely to be wrong — business name, coordinates, copyright — sit at the layer that changes least often. If they are set once during onboarding and never touched during production, an operator working fast cannot get them wrong. Platforms that support sub-accounts and per-client presets, which on SEOpix is the Agency plan, make this structural; without them you are enforcing it with a checklist, which works until someone is in a hurry.

Where a client has many locations, the coordinates are the field that needs the most care. Embedding a single head-office coordinate into images used across twelve branch pages is both inaccurate and pointless — the honest treatment of what geo-tagging images does and does not achieve for local SEO is worth reading before you decide how much effort to put into it, because the answer changes how you brief the work.

Briefs beat prompts

A prompt produces one image. A brief produces a consistent hundred.

For each client, write a short document — half a page is enough — that fixes the variables a prompt would otherwise leave to whoever is typing:

• Setting: interior or exterior, urban or suburban, the kind of premises
• Palette: two or three colours that match the site, and one to avoid
• Distance: wide establishing shots, mid-range working shots, or close detail
• Light: time of day and quality, which does more for consistency than any other single variable
• Subjects: what appears — equipment, premises, materials, process — and what never does
• Exclusions: competitor branding, readable signage, anything that dates quickly

Then every prompt for that account is the brief plus the specific subject. Two producers working on the same account in different weeks produce output that belongs together, which is the whole objective. The craft side of the prompt itself — getting a photograph rather than an illustration — is its own discipline and worth getting right once per agency rather than per client.

The delivery format question

Clients receive images in one of three ways, and the choice determines how much of your work survives contact with their site.

Raw files in a folder. Simplest, and the most likely to lose everything. A well-named file with correct EXIF uploaded through a CMS that renames and strips metadata arrives on the site as img_4471.jpg with no metadata at all. If this is your delivery route, tell the client's developer which platform behaviours to check.

Files plus a manifest. A spreadsheet mapping each file to its intended page, alt text and caption. This is the format that actually gets implemented correctly, because it tells the person uploading what the file is for. It costs about ten minutes per batch to produce and saves the round of "which one goes where" emails.

Direct upload by your team. Highest control, highest cost, and the only route where you can verify the final rendered markup. Reserve it for the client sites you also build.

Whichever route, the alt text is your responsibility, not the client's. It is the field most often left empty by the person doing the upload, and it is the one with an accessibility obligation attached. Writing alt text that is actually useful takes a few seconds per image once you know the rule for when to leave it empty. Put it in the manifest and it usually survives; leave it to the CMS and it usually does not.

Ownership, licensing and the conversation to have early

Two things are true at once and agencies routinely state only the convenient one.

The first: generated images can be used commercially under the terms of the platforms that produce them, and a client can put them on their site, in their ads and on their print collateral without a stock licence or a per-use fee. That is a genuine advantage over stock libraries with usage tiers.

The second: in the United States, an image generated from a text prompt is generally not itself protected by copyright, because there is no human authorship in the output. So the client cannot stop a competitor from producing something visually similar, and neither can you. The practical consequences — for exclusivity claims, for print, for anything a client might want to trademark — are set out in more detail in our piece on copyright and commercial use of AI images, and are worth understanding properly before you write a guarantee into a contract you cannot honour.

The clean position in a client agreement: the client receives an unrestricted commercial-use right to the delivered files, the agency makes no exclusivity representation, and anything requiring exclusivity — a logo, a trademarked mark, a signature product shot — is commissioned as original photography or design. That sentence has prevented more difficult conversations than any amount of careful hedging.

What to measure, and what not to

Agencies are asked to justify this line item, and the honest justification is not "images rank".

What images reliably do: fill pages that would otherwise be text-only, satisfy CMS templates that demand a hero, make a page look maintained rather than abandoned, and give the site something to serve into Google Images and visual search. What they do not reliably do on their own is move a page's position for a competitive commercial query.

So report the things that are actually attributable:

• Pages shipped with complete image sets versus pages still on placeholders
• Image impressions and clicks in Search Console, filtered to Search type: Image
• Largest Contentful Paint on templates where a hero image is the LCP element
• Cost per finished, deployed image, including your labour

That last line is the one that survives budget review, and it is the one most agencies never calculate. When you do, the tooling cost usually turns out to be a rounding error against the coordination time — which is the argument for spending effort on the workflow rather than on shopping for a cheaper generator.

A realistic onboarding sequence

For a new client account, in order:

1. Set the client sub-account fields once: business name, copyright, contact URL, coordinates per location.
2. Write the half-page visual brief with the account manager, and get the client to approve the brief, not the images.
3. Produce one batch of five to eight images and use them to calibrate. This is the only file-level review the client should do.
4. Lock the preset. Everything after this is production, not design.
5. Produce in batches keyed to page sets, delivering with a manifest each time.
6. Re-review the brief quarterly, or when the site's design changes.

Steps one through four take a couple of hours. Step five is where the volume happens and where it should feel boring — if batch production still requires creative decisions, the brief was not specific enough. The mechanics of running those batches efficiently, including the standardisation that has to happen before automation is worth anything, are covered in the bulk image SEO workflow.

The agencies that make this work are not the ones with better prompts. They are the ones who moved every recurring decision to the layer where it gets made once. See the Agency plan for how sub-accounts and white-label metadata fit that structure, or the feature list for what is included at each tier.