Skills

Ad variants in every placement size

Your agent composes each crop separately, with room left for the headline.

Meta, TikTok and YouTube crop differently, and one image scaled down loses the product or covers the copy. Your agent makes a version per size with the composition planned for that crop.

You say: "Build ad creative for our cold brew launch in feed, story and YouTube sizes."

Bildy tools this skill calls

• images_list_models
• images_generate
• files_upload
• images_edit
• images_get_job

The whole file

---
name: bildy-ad-variants
description: Make a batch of ad images from one product photo or one idea, across the sizes Meta, Google and TikTok ask for, each with room left for the headline. Use when someone needs creative for a campaign, several versions of the same ad, or one image resized for every placement.
---

# Ad variants from one picture

A campaign needs the same idea in several sizes, and each placement crops
differently, so a single image scaled down cuts off the product or covers the
headline. You make one version per size with the composition planned for that
crop.

## Sizes to cover

Ask which platforms they are buying. If they do not know, make these four:

- `1:1` for the feed
- `4:5` for the Instagram feed, which takes more vertical room
- `9:16` for Stories and TikTok
- `16:9` for YouTube and display

## Steps

1. Ask what the ad is selling and what the headline says. You need the
   headline to leave space for it, even though you are not putting the text in
   the picture.
2. If they have a product photo, upload it with `files_upload` and pass its
   `url` to `images_edit` in `imageUrls`. If the ad is
   an idea with no product, use `images_generate`.
3. Call `images_list_models` first and tell them what each image costs. Four
   sizes with two variations each is eight images, so say the total before you
   start.
4. Run one size, show it, and change the prompt until they are happy. Then run
   the remaining three with the same prompt and a different `ratio`.
5. Poll `images_get_job` until each job is completed, and hand over the
   download links. They are good for an hour, so save anything they want to
   keep.

## Writing the prompt

Put the copy space in the prompt. "The bottle in the lower third, the top half
empty sky, nothing in the upper area" gives a designer somewhere to put the
headline. Without it the model fills the frame and the headline ends up on top
of the product.

Describe the mood in things a camera can see. "Premium" means nothing to a
model. "Dark green backdrop, one hard light, deep shadows, the product wet
with condensation" produces the same feeling and you can picture it.

For a set that hangs together, keep the background, the light and the camera
angle identical across all four sizes and change only what fills the frame.

## Keeping the brand colour

Models drift on colour between runs. Name the colour in words the model knows
("a deep forest green", "a warm cream") instead of a hex code, and re-use the
exact same words in every prompt in the set. If the colour has to be exact,
generate on a neutral background and have a designer lay the colour behind it.

Other skills for the same tools