A product video clears brand, legal and marketing review. The media team uploads it, and the platform then makes a vertical crop, a shorter cut, an animated image or an entirely new video assembled from copy and product assets.
The operational question is no longer whether automation can create more versions. It is whether approval of one master gives the platform permission to create—and potentially apply—every derivative. Usually, it does not.
For brands connected to Chinese supply chains, this changes both the production brief and the acceptance process. Aspect ratios and deliverable counts are no longer enough. The handoff needs an explicit derivative-permission manifest shared by production, media, legal and the local market.
The signal: ad systems are moving from distribution to transformation
Google’s current Demand Gen documentation provides a concrete example. Video enhancements can create vertical, square and shorter versions from source videos, and the controls are enabled by default for new campaigns. Google also describes auto-generated video ads that use ad copy, images and brand guidelines to choose templates, colors, fonts and music. Adaptive image layouts, animated images and landing-page previews have their own default-on behavior.
Two other official sources show adjacent—but not identical—capabilities. OpenAI’s Ad Tools Terms, published August 24, 2026, say AI creative tools may generate, modify, transform, optimize, localize, translate, select, assemble or adapt creative; generated creative may be reviewed or automatically applied according to campaign settings, while the advertiser remains responsible for claims and disclosures. TikTok’s October 5, 2026 announcement describes Smart+ Creative Selection bringing brand assets, Symphony AI-generated creative and creator content into one approval workflow, with the advertiser controlling what goes live.
These sources do not prove that every platform behaves the same way or that the industry has completed a single transition. They do establish a procurement-relevant fact: the uploaded file can now be an input to further platform production, not necessarily the final output.
Master approval is not derivative permission
Approval of a master normally answers a narrow question: may this finished asset run in a defined market, channel and period? Derivative permission must answer four more:
- Does a new crop preserve the product action, warning and brand reveal?
- Does a shorter cut keep the claim and its proof together?
- Are new text, CTA, music or animation elements within the licensed scope?
- Is newly generated imagery faithful to the real product and reviewed for the destination market?
That distinction matters most for UI demonstrations, precise product operation, regulated or performance claims, price and availability, creator or talent footage, and music licensed for limited territories or uses. A version can look polished and still be unauthorized, misleading or commercially unusable.
The ONCE six-switch Derivative Permission Manifest
For every campaign, assign one of three states—Allowed, Review Required or Prohibited—to six transformation classes. “Platform adaptations permitted” is too vague to operate.
1. Reframe and crop
Allow it when the subject stays inside a documented safe zone and cropping cannot remove product details, captions or warnings. Require review for buttons, scales, ports or comparison views. Prohibit it where the crop would hide information needed to understand the claim.
2. Shorten and re-edit
Modular product shots and independent proof points are suitable candidates. Step-dependent demonstrations, disclaimers and claims that require context are not. A system can reduce duration; it cannot decide whether the remaining sequence is still truthful.
3. Add text and CTAs
Specify an approved copy library, prohibited terms, language, price validity and landing page. Do not give open-ended permission to rewrite performance, discount or comparative claims. Define visual exclusion zones so overlays do not cover the product or captions.
4. Animate still images
Packaging, material and lifestyle shots may be animated when motion cannot change product form. Review any movement that could imply a nonexistent mechanism, liquid behavior, transformation or use case.
5. Generate new visuals or video
Allow this only when the product truth source, reference assets and usage rights are explicit—and the team can inspect every proposed version. Precision components, medical or efficacy communication, real-person endorsements and unreleased packaging should not pass merely because the output looks plausible.
6. Create or apply the ad
Permission to generate is not permission to spend. The first five switches can be open while final application remains subject to human approval. Record the current platform settings, named approver, disable path and rollback asset.
Each switch also needs four fields: approved source or master, market and expiry, rights restrictions, and owner plus disable action. That turns permission into a production artifact rather than a meeting note.
Buy for safe derivation, not just final-file delivery
Production should anticipate transformations before the shoot. Capture safe compositions, clean plates, isolated product actions and separable audio. Post-production should deliver layered captions, text-safe zones, talent and music rights, claim sources, master version IDs and a snapshot of platform settings.
This extends an earlier ONCE principle: give every asset an identity before scaling AI ad creative. Asset identity answers “which file and who approved it?” The derivative manifest answers the next question: “what may the platform change after approval, under which conditions, and when must it stop?”
Automation is a better fit when claims are stable, shots are modular, rights are broad, safe zones are deliberate and review capacity is available. A fixed edit with manual adaptations is often more sensible for one-off launches, low-volume campaigns, tightly sequenced storytelling, regulated categories or complex rights.
Count approved derivatives, not generated outputs
Use a lifecycle denominator:
Lifecycle cost per approved derivative = (master and layered-asset production + permission configuration + platform generation/export + human QA/local review + rework and disable maintenance) ÷ derivatives actually approved and launched
Generated but rejected, disabled or never-launched variants do not belong in the denominator. This is not an industry benchmark; it is a management model. It prevents a misleading conclusion in which ten one-click outputs look inexpensive even when nine require rework.
Add one page to the next brief
Brands do not need to wait for platform behavior to converge. Add the six switches to the next video brief, select Allowed, Review Required or Prohibited for each, and attach the source, market, term, rights, approver and disable action. Save a pre-launch settings snapshot. Then approve only the derivatives that will actually run.
The purpose is not to slow automation. It is to show automation where it may move quickly—and where it must stop and ask.
Source note
This analysis uses Google Ads Help’s “About video enhancements in Demand Gen campaigns,” OpenAI’s “Ad Tools Terms” dated August 24, 2026, and TikTok Newsroom’s “What’s Next 2026: From Culture to Conversion” dated October 5, 2026. Each source establishes the named company’s product behavior or terms, not a completed industry-wide transformation. The six-switch manifest and lifecycle-cost model are ONCE frameworks, not platform rules or industry standards.
