Can GPT Image 2.6 make Facebook ads?

Yes. It can make a complete static ad, including the copy, product and button, from a prompt and reference images. Check every word and the product before you run it.

What size should I use for a Facebook ad with GPT Image 2.6?

We request 1088 x 1360 for a 4:5 feed ad and scale it to 1080 x 1350. For 1:1 we use 1088 x 1088, and for 9:16 1152 x 2048.

Can GPT Image 2.6 make a 1080 x 1350 image?

Not directly. Custom sizes need both sides divisible by 16, and 1080 is not. Request 1088 x 1360 and scale it down to 1080 x 1350.

Should I use Sunburst or Flare for ads?

Both made accurate ads in our tests at the same cost. Flare was two to three times faster and followed the reference layout more closely; OpenAI positions Sunburst for detailed work.

How much does a Facebook ad cost to make with GPT Image 2.6?

Our 4:5 ads with five reference images cost $0.088 to $0.108 each in OpenAI fees.

How to make Facebook ads with GPT Image 2.6

Updated 5 November 2026

To make a Facebook ad with GPT Image 2.6, send a proven ad as the layout reference, your product photos and logo, and a prompt that describes every zone of the new ad in your brand's words. Ask for Meta's exact ratio, such as 1088 x 1360 for a 4:5 feed ad, at high quality, then scale the result to 1080 x 1350 and check every word before it runs.

This is the process Mockingbird runs for every ad. Below are the steps, the settings that worked in our tests on 5 November 2026, and the faults to check for.

How to make a Facebook ad with GPT Image 2.6

1
Pick a proven ad as the reference
Choose an ad that has run for weeks in the Meta Ad Library. It gives the model a layout that already works, so you are only changing the brand, product and words.
2
Gather your product photos and logo
Use clear photos of the real packaging, up to four, and a logo file. Shrink anything larger than about 1536 pixels on the long side.
3
Write a zone-by-zone prompt
Describe the layout from top to bottom: each line of copy in quotes, the type style, the colours, where the product sits and the button. Say that the reference ad's brand, product and words must not appear.
4
Choose the model and settings
Use gpt-image-2.6-flare for speed or gpt-image-2.6-sunburst for OpenAI's detail model, quality high, and the size for your placement, such as 1088 x 1360 for 4:5.
5
Send the edit request
Send the reference ad first, then the logo and product photos, with your prompt, to the image edits endpoint.
6
Check and resize
Read every word, compare the product with your photos, then scale to Meta's pixels, for example 1080 x 1350.
7
Make the other sizes from the finished ad
Send the finished 4:5 ad back as the reference for the 1:1 and 9:16 versions, so the copy and product match across placements.

Settings that worked for us

Setting	What we use	Why
Model	Flare for speed, Sunburst for detail	Same cost per token; Flare was two to three times faster in our tests
Quality	high (xhigh and max also exist)	Clean small text and labels at a sensible cost
Size	1088 x 1360 for 4:5	Meta's exact ratio, just above the pixels you deliver
Complexity	Default (high) for new ads, low for small edits	OpenAI recommends high for dense text and layouts
Image order	Reference ad, logo, product photos	The model treats the first image as the main reference

Sizes for Meta placements

GPT Image 2.6 takes custom sizes with both sides divisible by 16, no side longer than 3,840 pixels and a shape between 1:3 and 3:1. Meta's 1080-pixel widths are not divisible by 16, so you cannot ask for 1080 x 1350 directly. Request a little above Meta's pixels and scale down, so the image is never stretched.

Placement	Request	Deliver
Feed, 4:5	1088 x 1360	1080 x 1350
Feed and carousel, 1:1	1088 x 1088	1080 x 1080
Stories and Reels, 9:16	1152 x 2048	1080 x 1920

For 9:16, tell the model to keep text out of the top 14% and the bottom 20% of the frame, where Instagram and Facebook place their own buttons. The full specs are in Meta ad specs and sizes.

How to write the prompt

The prompts that worked best for us describe the finished ad, not the idea. Go zone by zone and be literal.

Start with the format and size, for example "a 4:5 static Facebook ad".
Give every line of copy in quotes, exactly as it should read.
Name the type style from the brand's website, and the background and accent colours.
Say where the product goes and that it must match the attached photos exactly: packaging, label, shape and colour.
Describe the button and its text.
End with what must not appear: the reference brand's name, product and words, extra copy, star ratings you have not earned, and dashes between numbers.

Faults to check before an ad runs

Misspellings. Rare in our tests, but read every word anyway.
A redesigned product. Compare the label with your photos line by line.
Copy you did not ask for, such as invented reviews or badges.
Words or shapes carried over from the reference ad.
Text inside the 9:16 safe zones, where buttons will cover it.

Doing it without writing prompts

Mockingbird runs these steps for you. Pick a winning ad from a library approved by hand, and it reads your website for your brand voice, writes the prompt, renders the ad on GPT Image 2.6 with your real product photos and makes the other sizes from the finished ad. More on the approach is in how to make AI ads.

Frequently asked questions

Can GPT Image 2.6 make Facebook ads?: Yes. It can make a complete static ad, including the copy, product and button, from a prompt and reference images. Check every word and the product before you run it.
What size should I use for a Facebook ad with GPT Image 2.6?: We request 1088 x 1360 for a 4:5 feed ad and scale it to 1080 x 1350. For 1:1 we use 1088 x 1088, and for 9:16 1152 x 2048.
Can GPT Image 2.6 make a 1080 x 1350 image?: Not directly. Custom sizes need both sides divisible by 16, and 1080 is not. Request 1088 x 1360 and scale it down to 1080 x 1350.
Should I use Sunburst or Flare for ads?: Both made accurate ads in our tests at the same cost. Flare was two to three times faster and followed the reference layout more closely; OpenAI positions Sunburst for detailed work.
How much does a Facebook ad cost to make with GPT Image 2.6?: Our 4:5 ads with five reference images cost $0.088 to $0.108 each in OpenAI fees.