Meta, Google, and TikTok will generate your ads now. Should you let them?

On July 7, Meta shipped Muse Image, the first image model from its rebuilt Superintelligence Labs, and said advertisers will soon get access to it for marketing materials (Bloomberg). TikTok already offers Symphony Creative Studio to every logged-in TikTok for Business account (TikTok). Google Ads generates images from a prompt or straight from your landing page (Google). The three biggest ad platforms now all want to make your creative for you, inside their own walls, mostly for free.

That raises a question every marketing team will face this year: if the platform generates the ads, what is your creative pipeline for? The honest answer is that both sides win different jobs. Here is the verdict up front, then the detail.

The verdict

Use platform AI for last-mile adaptation: resizing, background swaps, quick variant tests, and formats native to a single platform. Keep master creative (the images, video, and copy that define how your brand looks) in a pipeline you own, where your brand kit, model choices, and review steps are pinned. If your ads are mostly product-feed retargeting and you have no brand system to protect, the platform tools alone may be enough, and you should use them without guilt. If your brand is the asset, treat platform generation as a distribution feature, never as the source of truth.

What each side actually is

What the platforms give you

The pitch is real. Google's asset generation reads your landing page and suggests images, headlines, and logos without a prompt; its image editor can feature your actual product in generated lifestyle scenes and accepts up to five style reference images to steer look and feel (Google). TikTok's Symphony Creative Studio builds TikTok-ready videos from a product URL or your existing assets, adds avatars, and translates and dubs voiceovers for other markets (TikTok). Meta's Muse Image will extend the same idea to the Advantage+ world: generate, alter, and vary ad images inside the apps where they run (Bloomberg).

Three advantages are hard to replicate. The tools are free or bundled, so testing ten variants costs nothing but review time. They are native, so output arrives pre-cropped and pre-formatted for the placement. And they sit next to the auction, which means the platform can (and will) generate variants tuned to what its own delivery system rewards.

Where platform generation breaks down

The weaknesses cluster around control.

One model, no substitutes. Each platform locks you to its own model. If Muse Image renders your product's material badly, or Google's generator can't hold your art direction, there is no swap. In a pipeline you own, a model is a node you replace when a better one ships.

Thin brand inputs. Five style references and a product photo are not a brand system. There is nowhere to pin your exact palette, your typography rules, your tone, or the list of things your brand never shows. Google's tools even restrict logos and branded items that aren't provided as input (Google), which is sensible for safety but limiting for identity work. Across dozens of variants, small deviations compound into ads that look like the platform's house style, not yours. The fix for that drift lives upstream, in reference images and a brand kit, not in per-asset prompt tweaks inside an ads manager.

Fragmented masters. Generate in Meta and the asset serves Meta. Generate in TikTok and it serves TikTok. Nothing connects them, so your "campaign" becomes three unrelated interpretations of the same brief. Google's generated images even expire from the Asset Library after 14 days unless used (Google). These are rendering surfaces, not asset systems.

Same tools, same output. Every competitor in your auction has access to the identical generator with the identical defaults. Whatever look the platform's model favors becomes the category's look. Differentiation has to come from somewhere the shared tool can't reach.

Everything is labeled. Meta watermarks Muse Image output invisibly, TikTok auto-labels every Symphony video as AI-generated, and Google embeds SynthID in every generated image. That's the right industry default, but it means the platform decides the disclosure posture for anything it generates. If you want control over how synthetic your brand reads, that's another decision that moves upstream.

Who should use which

Platform tools alone fit teams running catalog and feed-based ads where the product image is the creative, brands without a strict visual identity yet, and anyone testing whether a placement works before investing in real creative for it.

A pipeline first fits teams where brand consistency is commercially load-bearing: agencies answerable to client guidelines, ecommerce brands whose look is the moat, and any team producing one campaign across several platforms and markets. Build the master workflow once (brand kit in, generation steps, a human review gate) and every run produces assets that match, in the way the ad variants playbook lays out.

Most teams should run both, with a clean seam: masters from the pipeline, adaptation at the platform. Let Google resize and background-swap a master you made. Let TikTok dub the video your workflow produced. Feed the platforms finished creative and let their AI do placement-level work only.

How to try the split without a big migration

Start with one campaign. Produce the master set in a controlled workflow, on a canvas where the brand kit and review step are part of the graph, then export platform-neutral masters. Then enable the platform's generative features only for enhancement of those uploads, not net-new generation. Compare a platform-generated variant set against your pipeline set in the same auction. You'll learn quickly which placements reward native-generated creative and which punish the loss of brand signal, and you can draw the seam accordingly.

The platforms have made bad creative free. That changes the economics of testing, and you should take the gift. What it doesn't change is who owns the way your brand looks. That was never going to be Meta's model, or Google's, or TikTok's.