
Images Done for you by our team. Let's talk.
Your product stays your product.
Left alone, an image model redraws your product. Jinn cuts the real one out before anything renders, and every image is checked before you see it.
The starting point
A winning ad is a starting point, not a thing to copy.
Most AI ad tools hand you a blank prompt and hope. Jinn starts from an ad that is already running and staying up, because an ad that keeps running is the market telling you its shape works.
A vision pass keeps the part that transfers: the hook, the way it persuades, the layout and the call to action. Everything that belongs to the other brand is thrown away: its product, its name and every line of its copy. What comes back is not their ad recolored. It is your product and your claims, on a shape that has already earned attention.
Underneath it all sits your brand record, 700+ signals drawn from your own site and material. Jinn reads that record; it doesn’t make one up.
Every Images job runs one loop: your record, the shoot, the checks, then your yes.
Example brands, real Jinn output
Example brands, real Jinn output
You sign the set.
The checks report to you, not over you. You decide which frames go out.
Each finding arrives with the frame it was found on.
The build
A winning ad in, your ad out.
Four stages. Your price, offer and legal lines are set by fixed code, never drawn by a model. That is the difference between a prompt box and a process you can trust with a real campaign.
- Extract
A vision pass reads a proven ad and keeps only its reusable structure. It does not lift the other brand, its product or a single quote.
- Rebuild
That structure is rebuilt around your real product, cut out clean before any paid render starts, and around the claims your record can back. Exact and regulated copy stays out of the image prompt.
- Check
Every finished image runs a fixed quality pass. The contrast check has the final word on legibility, and it can send an otherwise clear ad back for a person to look at. Color is measured as a distance from your palette.
An AI vision judge also reviews each image, but only to advise: its notes reach you, and it never blocks a set or triggers a redo. What each check does when an image fails is in the table below.
- Ship
The set comes to you to approve and export. Each exported image is logged once, so a later measurement lands on the same record, and a real re-ship writes a new entry instead of overwriting the old one.
The checks
Each check, and what happens when an image fails it.
Some checks hold an image for a person. Others flag it and leave the call to you. Here is which is which.
| Check | What it looks for | If it fails |
|---|---|---|
| Product cut-out | A clean cut of your real product, made before any paid render | Your original photo is used instead of a broken cut-out |
| Exact words | Your price, offer and legal lines, set in your brand font after the render, never drawn by the model | With no brand font on file, the image is not made, so a legal line is never dropped |
| Logo check | A picture-reading pass for a rival’s logo or other protected mark | Held for a person: the image can’t be exported until someone accepts it. It is never re-created, and if the check can’t run, the image is held the same way |
| Garbled words | A picture-reading pass for broken or misspelled words in the image | One automatic retry. If the words are still garbled, the image comes to you flagged |
| Palette | How far the colors drift from your brand palette | A warning only. It doesn’t redo or block the image |
| Contrast | Whether every line of words is easy to read | Sent to a person to look at. The AI judge can’t overrule it |
Honest edges
Where the checks stop.
The honest read of any image process is where its checks stop. Here is where Jinn flags instead of enforcing, and what it doesn’t claim.
- The vision judge advises
A separate AI judge reviews each image and reports what it sees, but it doesn’t block a set or trigger a redo. Treat it as a second opinion, not a gate.
- Color is flagged, not enforced
The palette check measures how far an image drifts from your brand colors and warns when the reference ad’s look has survived. Today it flags a big miss; it doesn’t redo or block.
- Results don’t feed back
No loop runs from live ad results back into the images. Reference ads are ranked by how long they have been running, with no model involved, and exports are logged one way, for measurement only.
Questions
The questions people ask.
Is this just their ad with my logo dropped in?
No. Only the shape of the ad is kept; the other brand, its product and its copy are thrown out during the read. What gets built is your product and your claims on a proven shape.
Can the AI invent a stat or a fake review?
No. Only claims your record can back are eligible, and the exact copy is set into the image after it renders, so a number has nowhere to drift.
Does an AI decide whether an ad is good enough?
No. Each check has a set result: a bad cut-out falls back to your photo, a detected logo holds the image for a person, and low contrast sends it to a person to look at. The AI judge adds a second opinion and blocks nothing.

Images: want to see it run on your brand? Let's talk.
- Tell us about your brand.
- A person on our team reads it.
- We read every application and reply within two business days.
Already use Claude or ChatGPT? Connect your AI, free