Meta Muse: what Meta’s new AI models change for advertisers

What Meta announced

On 8 April 2026, Meta introduced Muse Spark, the first model in a new series built by Meta Superintelligence Labs. Meta describes it as a fast multimodal model that can reason and understand images. It powers the Meta AI app and is rolling out to WhatsApp, Instagram, Facebook, Messenger and smart glasses.

On 7 July 2026, Meta launched Muse Image, its first image generation model from the same lab. The announcement states that advertisers and agencies will be able to use it through Advantage+ creative.

On 23 September 2026, at its Connect conference, Meta put Muse at the centre of its strategy as an agent that can act for the user, with connectors open to developers and commerce partners.

What changes for ad creative

Advantage+ creative already offers generative tools: image expansion, backgrounds, text and animation of still images. At Cannes in June 2026, Meta also announced an end-to-end creative space and a “brand memory” that learns a brand’s identity and tone from its existing ads.

The consequence is simple: the cost of producing a variation drops sharply. The bottleneck moves to the brief: which angles to test, for which audiences, with which promises.

What still needs steering

Four decisions stay human, however good the model is:

• The brief: angles, objections to address and formats to test.
• Claims: a generated visual can promise what the offer does not guarantee, especially in finance, health or real estate.
• Brand consistency: a review before delivery, even a quick one, prevents drift.
• Measurement: a variation that gets more clicks only wins if it produces more qualified leads.

How to test safely

Start with a single ad set where generated variations run alongside creatives made by the team. Compare them on cost per qualified lead, not click-through rate, and note which angles win: that knowledge feeds the next brief.

Prepare your raw material

An image model starts from what it is given. The brands that get the most out of generative tools are those that organised their assets first:

• Sharp product photos, on a neutral background, from several angles.
• Colours, typefaces and logo in their official versions.
• Reusable proof: verifiable reviews, sourced figures, labels.
• The list of permitted claims, and the list of forbidden ones.

A brief that works with AI

The faster production gets, the more the brief matters. A useful brief fits in six lines: who you are talking to, which problem you address, which promise you make, which proof you show, in which format, and what must never be said or shown.

Write one brief per angle, not one brief per ad. The angle is what gets tested; variations of the same angle only refine it.

Measure the real contribution

To know whether generated creatives really help, compare them with the team’s creatives under the same conditions: same audience, same budget, same period, for at least two weeks. The criterion remains cost per qualified lead or per sale.

Watch fatigue too: variations mass-produced from the same visual look alike, and their performance can fall faster.

Risks to watch

Four points deserve a systematic check before delivery:

• Rights: faces, places and brands visible in the image.
• Claims in regulated sectors such as finance, health or real estate.
• Brand drift: colours, tone and style that slowly move away.
• Sameness: ads that end up looking like every competitor’s.

Questions