Anyone can generate an image. Shipping eleven that match is the job.
A working playbook for teams making pictures and footage with AI - not a gallery of prompts. How to brief a model like you'd brief a photographer, how to pick one without trusting a leaderboard, how to hold a look across a whole campaign, and how to be able to say where any of it came from afterwards.
- The six-slot prompt, with one brief written three ways
- A model bake-off you can run in an afternoon, and re-run quarterly
- 22 failure modes mapped to the specific edit that fixes each
- Eight copy-and-paste recipes, a one-page policy, and the record to keep
Four pages, unretouched
Screenshots of the actual PDF, generated from the same source - so what you see is what lands in your downloads folder.
Eleven chapters
and two appendices
Every chapter is one or two pages: a table you can act on, a worked example, or a checklist. Read it in ninety minutes, or raid the one chapter you need and come back later.
- 00 How to use this Who it's for, and three ways through it
- 01 What these tools are good at now Reliable, workable, still a demo - an honest sort
- 02 Picking a model without a leaderboard A bake-off you can run in an afternoon
- 03 The anatomy of an image prompt Six slots, and the same brief written three times
- 04 Consistency: brand, character, product The house style block, and the rule about your product
- 05 Video: think in shots, not clips Camera language, and why you generate the still first
- 06 The production workflow Brief, look dev, generate, select, finish
- 07 Failure modes and what fixes them 22 symptoms mapped to causes and specific fixes
- 08 Disclosure, provenance and rights What to mark, what to keep, and who tends to ask
- 09 A one-page policy you can adapt Green, amber, red - plus four sentences to fill in
- 10 First 30, 60, 90 days From someone experimenting to something normal
- - Appendix A · Recipe cards Eight prompts to paste, with notes on what to change
- - Appendix B · Glossary The twenty words people use loosely
Ideas that survive contact with a real brief
The six-slot shot brief
Subject, composition, light, environment, style, technical. Chapter 03 runs one real brief through three versions - from unusable to shippable - and annotates exactly what each edit bought.
An afternoon-long bake-off
Eight of your own briefs, blind scoring on five axes, and the metric that actually decides it: usable assets per hour, not price per image. Re-runnable every quarter, so the answer stays current.
Generate the scene, not the product
The rule that separates people shipping AI imagery commercially from people posting about it. A table of what to generate and what to composite, and how to ask for a plate you can drop a real photograph into.
The record, in eleven fields
What to capture at the moment you pick a keeper - prompt, model version, references, consent, disclosure decision, approver. Thirty seconds then; nearly impossible to reconstruct six months later.
“Generation is close to free and getting cheaper. That is not the interesting number. The interesting number is how many images you look at to find one you would put your name on - and looking is done by a person whose time is not free.”
Take the playbook. Share it with the team.
One PDF, no drip campaign, nothing to install. The chapter most teams wish they'd read first is 04 - the one explaining why their eleven campaign images don't look like they came from the same company.
Every image and video model your team uses, in one governed workspace.
The workflow in this book needs prompts and style blocks that live somewhere shared and versioned, the freedom to send each job to whichever model is genuinely best at it, and a record that writes itself. That is what StickyPrompts is for.