An AI Image Workflow for Agencies (That Actually Ships)

The three failure modes

Every agency that adds AI image generation without a workflow ends up in one of three bad places:

1. Everyone prompts differently → wildly inconsistent brand output
2. Nobody documents prompts → the good stuff can't be reproduced
3. Client sees the raw grid → cheapens the entire deliverable

Here's a workflow that avoids all three.

Roles

• Creative Director — approves the moodboard and brand style. Doesn't prompt.
• Prompt Engineer / Designer — writes the prompts, generates the batches.
• Curator — picks the top 3 from each batch. Usually the CD.
• Retoucher — 5–15 minutes of Photoshop polish per shipped image.

The stages

1. Brief (30 min) — Rough moodboard, brand palette, 3–5 must-have shots.

2. Prompt kit (1 hr) — Prompt Engineer builds a Brand Kit in Imagoat, drafts prompts for each must-have shot, generates 4 test images each.

3. Direction session (30 min) — CD reviews the tests, kills bad prompts, doubles down on good ones.

4. Generation (2–4 hrs) — Prompt Engineer generates 40–80 variations across the approved prompts.

5. Curation (1 hr) — Curator picks the top 15 %.

6. Retouch (2–4 hrs) — Light color grading, cleanup, resize, deliver.

Total: 1 day for a batch that used to take 2 weeks.

Documentation

Every shipped image needs a saved prompt in the project workspace. If a client asks for "the same style as that one from last month," you need to be able to reproduce it in 30 seconds.

Imagoat's Projects feature is designed exactly for this — every generation is tagged to a project, saved prompts are searchable, and Brand Kits carry the style suffix.

Pricing

Charge per deliverable, not per hour, and don't disclose that AI is part of the pipeline unless the client asks. The client is buying finished work, not knowledge of your process.