What changed
Three things happened to Meta's generative creative stack between July and September 2026. Discussed apart, they look minor. Together they change how an ad account behaves.
Meta put its own image model inside Advantage+ creative
On 7 July 2026, Meta announced Muse Image, the first image-generation model out of Meta Superintelligence Labs, initially in the Meta AI app alongside more than 30 new AI effects for Instagram Stories (Meta Newsroom). Three days later, on 10 July, Meta withdrew one launch feature — generating images by @-mentioning public Instagram accounts — after what it described as feedback that the feature "missed the mark" (Meta Newsroom). The rollout pace is fast enough that Meta is correcting in public, in days.
Advantage+ creative was named in the same announcement as a destination "in the coming weeks." Coverage at the time noted the advertiser integration was not yet live and that Muse Image would replace the model behind existing Advantage+ image work — background generation, lifestyle variations, and video-to-static conversion (The Keyword). Meta describes the model as using "agentic visual reasoning" and self-refinement, able to work from a creative brief rather than keywords, and to hold product appearance constant across variations (The Keyword). Those are Meta's characterisations, not measured results. Practitioner accounts put the Advantage+ rollout starting in July and widening through Q3 (Stephen Ellul).
The practical read: the generator that produces variants of your product photography got swapped out mid-quarter, under a setting most accounts leave on.
The opt-out surface is now per-feature, not one toggle
The single bundled "Standard Enhancements" switch was retired in early 2025. Every enhancement is now its own keyed entry inside a creative's creative_features_spec, usually nested under degrees_of_freedom_spec (Volume Creatives). Two consequences follow, and neither is cosmetic.
- Nothing can be turned off once and forgotten. Opt-outs are per feature, per creative. A policy set in March does not cover an ad built in September.
- New enhancement types default to Meta's choice. Meta keeps adding features; a deny-list written against last quarter's list silently admits this quarter's additions. Only an allow-list posture — explicitly opt out of everything known, re-check when the list grows — holds up.
There is a third point that matters more than either. Sending an opt-out is a request, not a guarantee. Features that are ineligible for a given creative are stripped from the spec without error, so what you submit and what is applied can differ (Volume Creatives). The only honest verification is to read the creative back after publish and compare it to what you sent. Counts of individual enhancements vary by source; one practitioner tally puts it above twenty (AdGenz), and a per-feature walkthrough covers image expansion, background generation, text generation, music, 3D animation, and filters (Flighted).
The default campaign is now the automated campaign
The agency Mintec, which says it manages more than 30 Meta ad accounts monthly, reports that as of early September 2026 Advantage+ became the starting state for new Sales, Leads, and App campaigns, with no accompanying announcement (Mintec). In their account, audience expansion, all placements including Threads and Audience Network, predictive budget allocation, and generative creative expansion all arrive pre-configured, and the exit control reads "Switch to original audience options" rather than anything resembling "disable automation" (Mintec). The same post states that generative creative expansion went generally available in August 2026 across North America and Western Europe.
Treat that as one agency's field observation, not a documented platform change. We could not find a Meta announcement confirming the September default flip or the August GA date. Meta's own Business Help Center page on Advantage+ creative does confirm the general principle — some enhancements are on by default, can be turned off at any time, and availability varies by audience and optimisation goal (Meta Business Help Center). Threads as a placement is independently corroborated as live in 2026 (Volume Creatives).
Why this matters to a brand spending $50K+/month
The commercial context explains the pace. Meta reported Q2 2026 ad revenue of $59.4 billion, up 27% year over year, on total revenue of $60.4 billion, up 28% — against expenses up 55% to $42 billion (Archon). On the same call, Meta said early tests of generative models in its ads retrieval system lifted clicks 8.3% and conversions 15.7% (Archon). Those two figures are Meta's own early-test numbers, self-reported on an earnings call — claims, not audited or independently replicated results, and they describe the retrieval and ranking layer rather than Advantage+ creative specifically. Meta's engineering blog describes the underlying ads foundation model, GEM, as LLM-inspired and trained across thousands of GPUs (Meta Engineering).
At $50K/month and above, four things break in a way they do not at $5K.
| What breaks | Why the spend level makes it worse |
|---|---|
| Creative A/B tests | Advantage+ reallocates budget and selects variant combinations per impression, so the two cells in your test are no longer comparable populations receiving comparable creative (AdLibrary). At low volume this is noise. At high volume it produces a confident wrong answer. |
| Creative attribution | If Meta generated the winning variant, your library does not contain the asset that won, so you cannot brief more of it. |
| Brand control | Music, text rewrites, background generation, and 3D animation apply per impression. Nobody signed off on the combination that ran. |
| Measurement continuity | Reporting definitions shift under existing metric names, so period-over-period comparisons that straddle a change compare two different things (Volume Creatives). |
The creative-testing problem is the expensive one. A brand at this spend level typically runs a weekly or biweekly creative cycle whose whole purpose is to learn which concept, hook, or format works. If the platform is silently varying the creative inside each cell and moving budget between cells, that loop stops producing knowledge and starts producing anecdote — while still costing the same in production.
What to do this month
1. Audit what is actually applied, not what you set
Pull every active creative via the API and read back creative_features_spec and degrees_of_freedom_spec. Compare the returned state to what your team believes it configured. Because ineligible features are dropped silently and opt-outs are requests rather than commitments (Volume Creatives), the returned state is the only reliable record. While you are in the account, check the account-level Advantage+ defaults and the "test new enhancements" option, which enrols creatives in enhancements you have not evaluated.
Output a spreadsheet: one row per active ad, one column per enhancement, applied versus intended. Expect gaps; the point is to size them before deciding anything.
2. Stop running creative tests inside Advantage+ campaigns
Where the objective is learning rather than short-term efficiency, run tests in a structure where you control the variables: manual placements, no audience expansion, enhancements explicitly off, budget fixed per cell. Accept that this cell may underperform on reported ROAS; that is the cost of an answer you can act on.
Keep Advantage+ running for the scaled spend. The split is not automation versus manual — it is a small measurement budget you can trust, funding decisions for a large delivery budget you cannot fully inspect. If your total is $50K/month, a $5–8K learning allocation is a reasonable place to start, and you should size it from your own variance, not from that ratio.
3. Write a one-page enhancement policy and encode it
Decide per enhancement, not per campaign, and write down the reason. A useful cut: enhancements that alter product appearance or generate net-new imagery are off for regulated, sized, or colour-critical categories; enhancements that only reframe or crop can stay on. Music and AI-generated text need a brand decision, once, rather than an argument every launch.
Then encode the policy in whatever creates your ads — API script, Supermetrics-style pipeline, or a documented manual checklist — and re-run the audit from action one monthly. Meta adds enhancement types faster than agencies update deny-lists; an allow-list posture is the only version of this that survives the next release.
What we would watch next
- A documented Muse Image availability notice for advertisers. The advertiser rollout is described in secondary coverage and practitioner reports, not a Meta advertiser-facing release we could locate. That is the biggest hole in this picture.
- Whether per-feature opt-outs survive the next release. The direction of travel since early 2025 has been toward fewer, broader defaults. If
creative_features_speccontrols are consolidated again, every account-level policy written this quarter needs rewriting. - Q3 2026 earnings, expected late October. Meta disclosed generative-AI lift figures for retrieval in Q2 (Archon). Whether it discloses anything comparable for generative creative — and whether the numbers are broken out or bundled — will indicate how confident Meta is in the creative side specifically.
- Incrementality methods becoming standard at this spend level. Geo holdouts and lift studies are less sensitive to in-platform variant selection than ad-set A/B splits.
What we could not verify
- The September 2026 default flip. Sourced to one agency post (Mintec) with no Meta announcement, no dated changelog entry, and no second independent report we could find. Their sample — self-described as 30+ accounts — may not generalise across regions or verticals.
- The August 2026 general-availability date for generative creative expansion in North America and Western Europe. Same single source; we found no corroboration.
- Muse Image's live status inside Advantage+ creative for all advertisers. Meta's July announcement said "coming weeks" (Meta Newsroom). Claims that the rollout completed across Q3 come from practitioner posts, not Meta.
- Meta's Business Help Center page on Advantage+ creative did not return readable content to our fetch tooling; we relied on its indexed summary. Read it yourself before treating our description of it as exact.
- Q2 2026 total revenue. One source reports $60.4 billion, up 28% (Archon); we saw a secondary figure near $60.8 billion elsewhere. Check the 10-Q before quoting either.
- The 8.3% click and 15.7% conversion lift figures. Meta's own early-test numbers from its own earnings call. No independent replication exists, and they describe ads retrieval, not Advantage+ creative.
- Enhancement counts. "More than 20" comes from a practitioner tally (AdGenz). Meta does not publish a canonical count, and the number visible in any given account varies by objective and placement.
The short version
Meta swapped the image model behind Advantage+ creative mid-quarter, the opt-out surface became per-feature and per-creative, and — by one credible but unconfirmed account — automation became the default entry point for new campaigns. None of this is a reason to abandon Advantage+; the delivery performance is real and the spend follows it. It is a reason to stop treating your ad account as a controlled environment. Read back what is applied, run learning in a structure you own, and write the policy down before the next enhancement ships.