Pilgud Agent now writes your headline into your own photo

Pilgud Agent now writes your ad's headline straight into your own photo – the picture of your team, your van, your finished job – instead of laying a text box on top of it. Then it reads the words back off the finished image, letter by letter, and won't let you approve an ad whose text came out wrong.

What's new?

What's new is that the words become part of the picture. Until now, an ad built on your photo was the photo with our text layer over it – readable, but plainly a caption placed on a picture.

Now, when the agent builds an ad on one of your photos, an image model draws the ad's lines into the photo itself: the headline first, then the subline and the offer line if the ad has them. The result reads like a poster made from your picture rather than a picture with a sticker on it.

The same goes for ads that don't start from a photo of yours. When the agent makes a picture from scratch, the headline is drawn in during the same generation, not added afterwards.

How does it work, step by step?

It all happens in the chat:

1. You add your photos. Attach them in the chat; they go into your image bank and the agent can use them straight away.
2. The agent writes the ads. Either it proposes variants built around your photo, or you describe an idea and point at the photos you want used. Copy is checked against the claims you've told it never to make, and variants the agent proposes go past its separate reviewer before they're saved – does it sound like a native speaker wrote it, does it argue one thing.
3. You get a draft at once. When you save the ads you want, each one appears immediately as a draft: your photo with our text layer. That draft costs nothing.
4. The headline is drawn in. The image model then writes the ad's words into your photo. It takes about half a minute, and the draft switches to the new picture when it's ready.
5. The words are read back. When the picture is ready, every line is read back off the image and compared with what the ad should say. If a single letter is wrong, the ad can't be approved. You can order a new picture, or keep your photo with our text layer instead.
6. You see a Facebook-style preview. Each ad is shown in chat as it will sit in the feed: your page name, the text above the picture, the picture itself, and the link bar with your domain, headline, description and the button the agent chose for that ad. The reviewer's score is shown with its reasoning.
7. You approve – and then you launch. Approving an ad doesn't put it live. When you build a campaign, it's created paused in your own Meta ad account, and nothing runs or spends until you switch it on with a click.

Because the words are now part of the picture, no ad shape may crop them. The agent makes a 4:5 version for the feed, which is the ratio Meta recommends for feed image ads, and a tall 9:16 version for Stories and Reels. In each, the whole picture is shown, on a background matched to the colour of your photo's edges.

Why does your own photo usually beat a stock or generated image?

For a local business, your own photo usually wins because it proves you exist and do the work. People buying a window clean or a bathroom renovation are asking one quiet question: are these people real, and are they any good? A photo of your actual work answers it before the headline does.

• It's recognisable. Your van, your street, your staff in your uniforms. Someone who has seen your truck in their neighbourhood stops scrolling for it.
• It's specific. A stock photo of a smiling cleaner could be anyone's.
• It's honest. A generated picture can show a kitchen you never built or a team you don't have. The agent won't pass a generated image off as your real work.

Writing the headline into the photo keeps all of that and adds what a caption can't: one poster-like image where picture and message are read together.

Example: a window cleaner's photo of a half-finished window – clean glass on one side, a grey film on the other – with "See the difference from the street" written across the clean half. The photo is the proof; the headline tells you where to look.

Who decides what runs?

You do. The agent drafts, draws and checks; you approve. An ad whose words read back wrong can't be approved at all. An ad you have already approved is never changed underneath you: if you ask for different words later, the agent makes a new version and draws the new words into it, and you approve that one.

Campaigns are built paused. Going live is your click, not the agent's. The same rule covers every ad Pilgud Agent makes – here's why we think an AI marketing agent must work this way.

What are the limits?

What it doesn't do, so nothing surprises you:

• A photo is raw material, not a finished ad. A dark, blurry or cluttered photo stays dark, blurry or cluttered with a headline on it. The agent treats your photo as raw material and may tell you it doesn't work as an ad.
• The image model adds the words; you check the rest. The aim is that only the words change and your photo stays as it was. Look at the preview before you approve – that's what it's for.
• Changing the words means a new picture. The words live in the pixels, so new text is drawn into a new version. The layout can't be adjusted separately, and if the drawn version fails or its words come back wrong, you can switch back to your photo with our text layer.
• Other photo edits aren't available. Retouching, removing objects or restyling your photo isn't something the agent does.
• Each drawn picture has a small generation cost and counts towards your monthly picture budget. If the drawing can't start, you keep the free draft.
• Stills first. This is for image ads; for motion, the agent makes short video ads from your photos separately.

To try it, open Pilgud Agent, attach a few photos of your work and ask for ads. Plans are on the pricing page.