If You Turned Off Advantage+ Creative Because It Looked Bad, You Tested a Model That No Longer Exists

If you turned off Advantage+ Creative because the AI variations looked bad, you tested a version that doesn’t even exist anymore.

Meta just swapped the entire model powering those AI creative enhancements.

It’s called Muse Image, their first image model built by Meta Superintelligence Labs, and it’s rolling into Advantage+ Creative across advertiser accounts this quarter.

What Actually Launched

Meta released Muse Image on July 7, 2026, the first image generation model built entirely in-house by Meta Superintelligence Labs, the research group led by Alexandr Wang.

It is the second major release from that team after Muse Spark, the large language model that replaced Meta’s open-weight Llama lineup in April, according to Meta’s own announcement.

Until now, Meta had leaned on outside vendors, including Midjourney and Black Forest Labs, to power image generation inside its apps.

Muse Image ends that dependency for the first time.

This is not a minor model refresh. Muse Image uses native reasoning to work through a prompt before generating anything, allowing it to plan layouts, blend multiple source photos together, and pull in real-time web context when relevant, according to Neowin’s coverage of the technical details.

It builds directly on Muse Spark to better interpret complex prompts and visual references, which is a structurally different approach than the third-party models it replaces.

Where It’s Headed: Advantage+ Creative

Meta’s own business-facing announcement is direct about where this goes next: advertisers and agencies will be able to tap into Muse Image through Advantage+ Creative in the coming weeks, powering background generation and expansion, image touch-up, text generation, and image variations directly on top of an advertiser’s existing assets, according to CNBC’s reporting on the advertiser rollout.

Meta describes the goal as bringing native reasoning to the creative process so the model can adjust elements, swap styles, and generate on-brand variations with fewer iterations than the tools it replaces.

This is not a new feature being bolted onto Advantage+ Creative.

It is a swap of the engine sitting underneath a feature advertisers already know, one that has been generating backgrounds, image edits, and creative variations for accounts that had it turned on, according to Digital Applied’s paid social playbook for the rollout.

Why the Old Test You Ran Doesn’t Apply Anymore

Here is the part worth sitting with.

If your team ran a test back in the spring, hated the output, and quietly turned the feature off, that decision was based on a model Meta has now completely replaced.

The generation quality baseline just moved.

Whatever conclusion you reached about AI creative on Meta a few months ago might already be wrong, not because your judgment was bad, but because the thing you judged no longer exists in the form you tested it.

This distinction matters more than it sounds like it should.

A feature name staying the same while the underlying model changes underneath it is exactly the kind of update that never shows up as a headline anyone reads before making an account-level decision.

Most teams that disabled Advantage+ Creative months ago have no reason to know the model swap happened at all unless they are specifically watching Meta’s product announcements.

How Good Is It, Actually

Worth being honest about where Muse Image actually sits rather than repeating Meta’s own launch framing uncritically.

Based on Meta’s own benchmarks, Muse Image trails OpenAI’s GPT Image 2 overall, but performs better than Google’s Nano Banana 2 specifically on single and multi-image editing tasks, and ranks second on the Arena leaderboard, according to Forbes’ analysis of the launch.

That is a genuinely competitive position, not a category leader claim.

It is also specifically strong at the editing task type that Advantage+ Creative actually leans on, since most of what the feature does is adjusting and varying existing advertiser assets rather than generating images from a blank prompt.

What This Means Structurally

Muse Image is part of a broader pattern of Meta bringing creative production in-house rather than licensing it.

Pairing an internally built, reasoning-capable image model with Advantage+ Creative means Meta now controls both the delivery algorithm and a meaningfully larger share of the creative pipeline feeding it, rather than relying on a third-party model whose quality and roadmap Meta did not control.

That matters for advertisers because the two systems, creative generation and delivery optimization, are increasingly built by the same team with the same incentives, not stitched together from separate vendors.

It also means future quality changes are now entirely inside Meta’s control, for better or worse, rather than dependent on a licensing relationship with an outside vendor that could change terms or access independently of anything Meta itself decides.

The Compliance Layer Worth Knowing About

Content generated through Muse Image inside Advantage+ Creative is still subject to Meta’s AI-generated content labeling requirements and Content Seal watermarking, which govern what actually ships inside Ads Manager and how it gets disclosed.

If your team re-tests the feature, build in a check on how labeling and watermarking apply to the specific asset types you generate, since compliance requirements sit on top of the creative output itself and do not change just because the underlying model did.

What AdBeacon Has Covered on AI Creative

AdBeacon has written about the tradeoffs of AI-generated creative more broadly in AI creative wins clicks, loses conversions once AOV crosses $100, and about how brands are scaling AI creative production in AI creative generation and UGC scaling for ecommerce.

Both are worth revisiting with this specific model swap in mind, since the underlying performance patterns those pieces describe were measured against a different generation engine than the one now live.

What To Actually Do This Week

Go back into Advantage+ Creative this week and re-test with fresh eyes, especially if you wrote it off before.

Do not assume your spring test result still holds. Run a clean, small-scale test against current output before making any decision based on an evaluation that predates July 7.

Compare a handful of Muse Image-generated variations directly against your original creative, not against your memory of what the old model produced.

Check how AI-generated content labeling and Content Seal watermarking apply to whatever asset types you generate, since that governs what actually ships, not just whether the output looks good.

If you manage multiple accounts, note that rollout is happening progressively across advertisers this quarter, so a re-test that shows nothing changed yet is worth repeating again in a few weeks rather than treated as a final answer.

Document the specific asset types you re-test, image touch-up, background expansion, or full variations, since Meta’s own benchmarks suggest editing tasks are where this model is strongest relative to competitors.

None of this means Muse Image will work for every brand.

It means the version you judged and rejected is not the version currently running, and that gap is worth closing before you write off the feature again.

If you want to see how AI-generated creative variations actually perform against your independently measured, click-based conversion data, book a live AdBeacon demo and look at the real numbers.

FAQ

What is Meta’s Muse Image model?

Muse Image is the first in-house image generation model built by Meta Superintelligence Labs, released July 7, 2026, replacing Meta’s prior reliance on third-party models like Midjourney and Black Forest Labs for image generation across its apps and advertising tools.

Does Muse Image power Advantage+ Creative now?

Meta announced that advertisers and agencies would gain access to Muse Image through Advantage+ Creative in the coming weeks following the July 7, 2026 launch, with the rollout continuing across advertiser accounts this quarter.

How does Muse Image compare to other AI image models?

Based on Meta’s own benchmarks, Muse Image trails OpenAI’s GPT Image 2 overall but outperforms Google’s Nano Banana 2 on single and multi-image editing tasks, and ranks second on the Arena leaderboard.

If I turned off Advantage+ Creative months ago, should I reconsider?

Yes, it is worth re-testing. If your evaluation happened before July 7, 2026, you tested output from a different, third-party-powered model than the one currently rolling out, so that earlier conclusion may no longer reflect current quality.

Does AI-generated content from Advantage+ Creative need to be labeled?

Yes. Content generated through Muse Image inside Advantage+ Creative remains subject to Meta’s AI-generated content labeling requirements and Content Seal watermarking, which govern how the content is disclosed inside Ads Manager.