Meta Andromeda Creative Strategy: 20 Ads From One Product Photo

What is Meta Andromeda?

Andromeda is Meta's personalized ads retrieval engine: the stage that shortlists which ads are even considered for a person before the auction runs. When someone opens Facebook or Instagram, the system narrows tens of millions of ad candidates down to a few thousand. Ranking and the auction then happen on that shortlist. Andromeda builds it.

Meta's engineering team announced the system on December 2, 2024. According to that post, Andromeda runs a deep neural network on NVIDIA Grace Hopper Superchips and Meta's own MTIA accelerators, and was already deployed across Instagram and Facebook at the time. Advertisers only started talking about its effects widely during 2025, which is why you'll often see it discussed as "the Andromeda update."

All four figures come from Meta's own engineering blog. What they mean for you is simple: the system now reads the content of your ad (the image, the copy, the video) in far more detail, and leans heavily on that content to decide who should see it.

Why did Andromeda change Facebook ad strategy?

Because the weight of targeting shifted from audience settings to the creative. The person you used to describe as "women, 25-34, interested in running" is now described by the scene in your ad, who is in it and what it says.

Meta's push toward broad targeting and Advantage+ audience fits this. Rather than narrowing the audience, you give the system room and let the signals inside the creative steer delivery. An ad showing a running shoe on a 45-year-old woman out on a rainy morning run will reach different people than the same shoe on a white studio background. Put both in one campaign and the system has two separate "audience keys" to work with.

Why are similar creatives treated as one ad?

Meta says ads that look too alike come across as repetitive, and practitioners consistently see such ads compete as one idea and share delivery. Media buyers often call this grouping the "Entity ID." That's industry shorthand rather than a documented Meta setting, and Meta hasn't published exactly how the grouping works, so treat any "similarity score" thresholds you read online as rules of thumb. Meta's December 2025 post on creative diversification separates two ideas: creative fatigue, when people see the same ad too many times, and creative similarity, when different ads look so alike that people perceive them as repetitive.

In practice: run one product photo with five background colors and three headlines and you have 15 ads, but to the system most of them are one concept. It's also the usual reason new ads get no spend at all. The system sees them as copies of an idea it has already tested.

You can track fatigue and similarity in Ads Manager under Analyze & Report → Ads Reporting → Account Insights, in the creative insights section. Menu names vary by interface language and account, and as of September 2026 these insights haven't reached every account at the same time.

How many ads should you have per ad set?

The hard limit is 50 ads per ad set; the right number is however many distinct concepts your budget can fund to a meaningful level of spend. Meta's November 2024 guidance for Advantage+ shopping campaigns (now called Advantage+ sales campaigns) is to run at least 20 diversified ads and put 20-30% of total budget toward testing new ads.

That 20 means "20 different ideas," not "upload 20 files." For a smaller store, a simple decision rule helps:

Worked example: your daily budget is $300 and your target CPA is $30, so the ratio is 10. That account can carry roughly 10-12 concepts at once. If you set aside 25% for testing, that's $75 a day, or $525 a week. Launch 5 new concepts a week and each one gets about $105 in its first week, roughly 3.5 CPAs. Not judging a concept until it has spent at least 2-3 CPAs cuts down on killing ideas too early. This table is not an official Meta threshold; it's a starting rule to tune with your own account data.

How do you turn one product photo into 20 different ads?

Multiply the story, not the photo: who it's for (persona), in what situation (context), what convinces them (proof) and in which format. AI image tools now let you place a product in new scenes, swap backgrounds and export different aspect ratios in minutes. We explain why a clean product shot is the right starting point in our guide to AI product photography.

1. Prepare one clean hero image A single high-resolution product photo with even lighting and the background removed.
2. Write 3-4 personas Who buys this, and why? For example: new parent, student athlete, gift shopper, price-conscious buyer.
3. Pick a problem and a scene for each Create a realistic setting that places the product inside that person's day.
4. Add a proof layer A review, comparison, guarantee, material close-up or other verifiable reason to believe.
5. Change the format Build the same idea separately as a single image, a 9:16 Reel, a carousel and a short text-led banner.

Here is a sample 20-concept matrix for an insulated travel mug. Each row targets a different person with a different core message, which is what lets the system read them as separate opportunities.

Meta's own Advantage+ creative tools also generate images and text, and in its December 2025 post Meta reported 11% higher CTR and 7.6% higher conversion rate for campaigns using its image generation feature. Still, these tools mostly produce variations of the same concept. The decision about different personas and contexts is yours to make.

How do you spot creative fatigue, and when should you refresh?

As the same audience sees an ad more often, cost per result climbs, and Ads Manager can flag this as a delivery status on the ad or ad set. According to Meta's Help Center, for eligible ad sets the Delivery column shows "Creative limited" when cost per result is above your past ads but less than double, and "Creative fatigue" when it is double or more. Even so, fatigue rarely shows up in one metric; look for several signals moving together.

• Frequency is rising while CTR has fallen two weeks in a row.
• Cost per result is clearly above the ad's own historical average.
• New ads get no spend; budget keeps flowing to the same one or two old ads.
• Account Insights shows a creative similarity warning.
• Comments like "I keep seeing this ad" are increasing.

When you refresh, don't clone the winner and change its color. Work out why the winning concept worked, then rebuild that reason for a new persona or context. If the leak test performed well, for instance, carry the same proof into an unboxing video for gift shoppers. For how costs move over the year, see our breakdown of Instagram ad costs in 2026.

How should you structure campaigns after Andromeda?

Fewer campaigns, fewer ad sets, more concepts: simplify the structure and keep the diversity at the creative level. A separate ad set per persona splits budget and slows learning, and the system can already read the persona from the creative.

• Main sales campaign: one broad Advantage+ sales campaign or a single broad ad set where proven concepts run. For setup details, see our Advantage+ sales campaigns guide.
• Testing space: a separate campaign or ad set with 20-30% of budget where new concepts are tried. Winners graduate to the main campaign.
• Measurement: run the Pixel and Conversions API together, or the system can't reliably learn which concept drives sales. Our Conversions API setup guide walks through it.

In your weekly review, focus on one question: which concept brought in sales, and at what cost? That gets hard to answer when Meta, GA4 and your store numbers disagree. Marpany's ad performance view brings channels into one table so concept decisions rest on the same data, and you can read how those decisions can be automated in our article on AI-driven ad management.

What should you do this week?

Count your current ads by concept first, then build new ideas for the personas you're missing.

1. List your active ads and mark how many genuinely different concepts you have. Count color and headline variations as one.
2. Work out your daily budget / target CPA ratio and set a target concept count from the table above.
3. Write 3-4 personas for your best-selling product, and pick at least one scene and one proof point for each.
4. Generate those scenes from a single product photo with AI, and build every idea in at least two formats.
5. Put 20-30% of budget into a testing space and launch new concepts on the same day each week.