Skip to content
AI stack, job

AI for static ad creatives

Last verified

Start a static ad from the concept, then pick the image model. Test candidates on your own product photo and on one line of text inside the image, since rendered text and product accuracy fail most often. Check commercial-use terms and platform AI labels, then judge winners on cost per result in a live test.

The workflow for this job
The workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESConcepts andimage drafts02 CHECKRights, AI label,legibility03 PERSON APPROVESPicks thecreatives04 SHIP AND MEASURELive test, killruleNEXT BRIEFThe workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESConcepts and image drafts02 CHECKRights, AI label, legibility03 PERSON APPROVESPicks the creatives04 SHIP AND MEASURELive test, kill rule
Illustrative flow, simplified.
The model proposes, a check catches what it got wrong, a person approves, and the result ships and is measured. What you learn feeds the next brief.

Concept first, model second

A static ad is a message, a visual and a reason to tap. Write the concept before you open any image tool: the audience, the single claim, the proof and the offer. A model can render a weak idea in high resolution, but the ad will still fail.

Turn the concept into a brief the model can follow. List the product, the scene, the layout, the one line of text that goes inside the image, and the formats you need. The same brief then goes to every candidate model, so the comparison is fair.

Image model options

Read the table below by job, host and price source. The checked date matters, because vendors change prices and hosts such as Runway, Luma and Higgsfield resell other vendors' models at their own credit prices. The table says so when a price is not verified, instead of guessing.

Models for static ad creatives
NameVendorJobStatusPrice as the vendor states itCheckedSource
GPT ImageOpenAIStatic creativesCurrentSee vendorOpenAI
Nano Banana family (Gemini image models)GoogleStatic creativesCurrentPer image: Nano Banana 2.1 about $0.034 (1K), $0.050 (2K), $0.113 (4K); Nano Banana 2 $0.067 (1K); Lite about $0.034 (1K); Pro $0.134 (1K/2K), $0.24 (4K).GoogleGoogle
Grok ImaginexAIStatic creatives, Video and UGCCurrentSee vendorxAI
Advantage+ creative generative AI features in Ads ManagerMetaStatic creatives, Ad copy, Market researchCurrentSee vendorabout.fb.com
Qwen models (Model Studio)AlibabaAd copy, Market research, Static creatives, AnalyticsCurrentSee vendorAlibaba
FLUX 3 and FLUX.2Black Forest LabsStatic creativesCurrentSee vendorBlack Forest Labs
Runway Gen-4.5 and Aleph 2.0RunwayVideo and UGC, Static creativesCurrentPlans: Free $0 (125 one-time credits), Standard $12/mo annual ($15 monthly, 625 credits), Pro $28 ($35, 2,250), Max $76 ($95, 9,500).Runwayrunway.com
Ray3.2 and Ray3.14Luma AIVideo and UGC, Static creativesCurrentPlus $30/mo ($25 annual) 10,000 credits, includes commercial use; Pro $90 ($75) 40,000; Ultra $300 ($250) 150,000; Team and Enterprise by sales.Luma AILuma AI
HiggsfieldHiggsfieldVideo and UGC, Static creativesCurrentSee vendorHiggsfield

There is no clear leader for image creatives. Nano Banana, GPT Image, FLUX, Grok Imagine and Meta's Advantage+ generative AI tools overlap, and the right pick depends on your product, your text and your cost per result.

Write the model ID as the vendor writes it. OpenAI lists gpt-image-2.5 and gpt-image-2.6, and its image prices are quoted per token, so a per-image figure needs the arithmetic shown. For Nano Banana 2.1, Google lists about $0.034 for a 1K image, and it also lists price increases on 1 December 2026 and 1 January 2027.

To estimate generation cost, multiply the per-image price by the number of variants you plan to review. Then add review time and rejected outputs, since the image fee can be small next to media spend and the hours your team spends checking outputs.

Meta's Advantage+ creative tools live inside Ads Manager and cover image variations, image expansion, text overlays and copy variations. The current 2026 feature list was not verified, and no official 2026 product page was found, so open Ads Manager and check what your account shows.

Test on your product photo and one line of text

Rendered text and product accuracy fail most often. Test each candidate on your own product photo and on one line of text inside the image, such as the headline you plan to run.

  1. 1.Fix the inputs

    Pick one real product photo, one brief and one text line. Use the same three inputs for every model and host.

  2. 2.Generate a small batch

    Make the same number of outputs per candidate, with no hand-picking before the review.

    Brief: product, scene, layout. Text in image: your headline line. Format: 1:1 and 4:5.
  3. 3.Score against the real photo

    Compare shape, colour, label and logo with the original. Check the text for spelling, spacing and legibility at phone size.

  4. 4.Record cost and effort

    Write down price per output, number of retries and minutes of editing. Name the host, since credit prices differ by host.

  5. 5.Keep two finalists

    Take the outputs that passed your checks from two candidates into the live test. A winner chosen on looks alone is a guess.

Pass and fail checks before an image reaches the ad account
CheckWhat to look atReject when
Product accuracyShape, colour, packaging and logo against the real photoAny detail differs from the product you sell
Text in imageSpelling, spacing, contrast at phone sizeA letter is wrong or the line is hard to read
Hands and facesFingers, eyes, teeth and skin at full zoomVisible artefacts a viewer would notice in a feed
ClaimsAny badge, number or promise the model addedThe image states something you cannot back up
FormatCrop safety for 1:1 and 4:5 placementsKey text or product sits in a cropped zone

Before an image goes to the ad account, check the words that ship with it: primary text, headline and button.

Pre-flight your static ad text

Commercial use and AI labels

Commercial rights differ by vendor and by plan. Among the hosts checked, Luma lists commercial use on its Plus plan. Runway's pricing page did not state commercial terms. The commercial terms for the image models above were not verified here, so read the vendor terms for the exact plan you pay for before a client ad goes live.

Meta labels ads made with its own gen AI tools as AI info, under a policy updated on 1 June 2026. Whether images from third-party models trigger a label or other platform rules was not verified. Check Meta's current ad policy before launch, and keep a record of which tool produced each asset.

Judge in a live test

Looks do not decide winners. Cost per result in a live Meta test does. Set the budget, the audience and the winner and kill criteria before launch, so the numbers decide and not the opinions in the room.

The first-party ua5er/user-acquisition-skills pack includes meta-creative-testing. It designs creative testing for Meta, picks which past ads to relaunch, sets winner and kill criteria and writes briefs for designers and UGC creators. Use its kill rules as a starting template and adjust them to your margin.

Skills for creatives

The tables below list third-party skill packs and an MCP server for static creatives. Treat each one as unreviewed code: skills may bundle scripts, and licences differ by repository. Check the licence and read the scripts before you run or host anything.

Skills for this job
NameJobLicenceChecked
JimLiu/baoyu-skillsStatic creativesMITClass A
mbfinotti/advertising-skillsStatic creativesMITClass A
MCP servers for this job
NameJobPublisherAccessLicenceChecked
pipeboard-co/meta-ads-mcpPaid UA agents, Analytics, Static creativesThird partyNot reviewedBUSL-1.1Class B

What this page does not claim

  • This page does not rank image models. The checks behind it gave no basis for picking one over another, so the choice stays with your own test on your own brief.
  • Prices are shown only where the vendor page gave them, with a checked date. Midjourney and Seedream prices were not verified.
  • Whether third-party AI images trigger platform labels was not verified.
  • Third-party skill and MCP repositories are listed for discovery. They are not endorsed, and the safety scans were file lists plus pattern search, with nothing executed.

Questions

Questions

Which AI image generator is best for Meta ads?
No clear leader. Nano Banana, GPT Image, FLUX and Grok Imagine all compete, and results depend on your product and your text. Run the same brief and product photo through two or three of them, then compare cost per result in a live Meta test.
Can AI make product photos for ads?
It can build scenes and backgrounds and edit from a reference photo. Product accuracy is a frequent failure, including shape, label text and logo. Compare each output with the real photo, reject any that alter the product, and keep the unedited photo as a fallback.
Do AI-generated ads need a label on Meta?
Meta labels ads made with its own gen AI tools as AI info, under a policy updated on 1 June 2026. Whether images from third-party models trigger a label or other rules was not verified here, so check Meta's current policy before launch.
What does an AI image cost per ad?
It depends on the model and host. Google lists about $0.034 for a 1K image on Nano Banana 2.1, checked 15 November 2026, with increases listed for 1 December 2026 and 1 January 2027. OpenAI quotes image prices per token, and a per-image price was not verified.

In one real market, 12 of 20 static ads named an offer and 10 cited social proof. See the run, the briefs it led to and the panel that screened the concepts.

See the real run

Sources

Each source shows the date it was last checked.

Show all 19 sources
  1. about.fb.com about.fb.com/news/2025/02/gen-ai-transparency-metas-ads-products/
  2. Google aistudio.google.com/docs/models
  3. Google aistudio.google.com/docs/pricing
  4. Alibaba alibabacloud.com/help/en/model-studio/models
  5. Black Forest Labs bfl.ai/pricing
  6. OpenAI developers.openai.com/api/docs/models
  7. OpenAI developers.openai.com/api/docs/pricing
  8. Black Forest Labs docs.bfl.ai
  9. Runway docs.dev.runwayml.com/guides/pricing/
  10. xAI docs.x.ai/docs/models
  11. Meta facebook.com/business/news/Introducing-Enhanced-Gen-AI-Features-and-Other-Tools-to-Help-Build-Your-Business
  12. GitHub github.com/JimLiu/baoyu-skills
  13. GitHub github.com/mbfinotti/advertising-skills
  14. GitHub github.com/pipeboard-co/meta-ads-mcp
  15. Higgsfield higgsfield.ai/blog/higgsfield-pricing-explained
  16. GitHub ua5er/user-acquisition-skills: LICENSE
  17. Luma AI lumalabs.ai/pricing
  18. GitHub ua5er/user-acquisition-skills
  19. runway.com runway.com/pricing

Need creative testing and AI tooling tied to your app's cost per install? Start with a short audit.

See the service