AI in Retail Marketing: Legal Risks of Using Synthetic Images, Videos, and CGI

Here’s a practical, jurisdiction-by-jurisdiction briefing for retail leaders who are considering AI-generated images, videos, and CGI/3D animations across ads, product pages, packaging, in-store screens, and social. It focuses on what’s legally risky, what’s required, and what good looks like in day-to-day workflows. I’ve grouped the rules into five major markets—the United States, European Union, United Kingdom, China, and Canada—then closed with a universal compliance playbook and sample policy language you can adapt.

Executive summary

  • Truthfulness beats labels. In every market, the core rule is the same: don’t mislead consumers. If an AI/CGI visual materially changes how a product looks, works, or performs, you must fix the visual or add clear, conspicuous disclosures—often both. (US FTC Endorsement Guides & deceptive practices; EU UCPD & DSA). European Commission+3Federal Register+3Federal Trade Commission+3

  • Human vs. machine authorship matters for copyright. Purely machine-generated output often isn’t copyrightable (US); mixed human/AI works can be, if you document the human creative contribution. Plan licensing accordingly. AP News+3U.S. Copyright Office+3U.S. Copyright Office+3

  • Faces and voices = high risk. Using a person’s likeness (realistic avatars, voice clones) requires consent under right-of-publicity and privacy/biometrics rules; misuse triggers quick enforcement and reputational damage. Justia Law+2dwt.com+2

  • EU AI Act adds transparency duties for synthetic media (“deepfakes”). If you deploy AI systems to create synthetic content, you must disclose that the content is AI-generated or manipulated (with limited exceptions). Start building labeling and watermarking into your tooling now. EUR-Lex+2White & Case+2

  • Platform obligations affect you even if you’re not the platform. The EU Digital Services Act requires ad transparency and restricts targeting; major platforms enforce these rules against advertisers. Assume your creatives will be audited. European Commission+2European Commission+2

  • China requires labeling/watermarking and places duties on providers of “deep synthesis.” If you localize creative for Mainland China, integrate labeling, watermarking, provenance records, and additional content controls. China Law Translate+2Carnegie Endowment+2

  • Keep audit trails. Save prompts, edit histories, training/asset licenses, and QA sign-offs. If a regulator asks “How did you make this visual?” you need a fast, credible answer.

United States (US)

Deceptive or misleading visuals

The US doesn’t require generic “AI used” labels across the board, but it does aggressively police misleading ads and claims. If your AI/CGI depiction misrepresents the product (e.g., a 3D render that hides a size limitation, digitally “improves” skin results, or implies features that don’t exist), you risk enforcement under Section 5 of the FTC Act. The FTC’s recent actions emphasize there’s no AI exemption to advertising law; “AI-powered” claims also need substantiation. Use disclosures when visuals could otherwise mislead, and ensure performance claims are backed by evidence. Federal Trade Commission+1

Influencer/UGC integrations. If creators post AI-stylized looks or virtual try-ons, material connections must be disclosed under the updated 2023 Endorsement Guides (e.g., “#ad,” prominent and unavoidable; platform tags alone may be insufficient). Build this into briefs and monitoring. Federal Register+1

Right of publicity & likeness

Using a person’s name, image, likeness, or voice for commercial purposes without consent can violate state right-of-publicity laws (e.g., California Civil Code §3344; New York Civil Rights Law §50-f, including “digital replicas” and post-mortem rights). AI-generated doubles and voice clones are squarely in scope. Maintain signed releases and talent contracts that cover digital replicas, voice synthesis, model retraining, and future media. Justia Law+2FindLaw Codes+2

Copyright & ownership of AI outputs

The US Copyright Office rejects registration for purely machine-generated works; however, works with sufficient human creativity can be protected, and applicants must disclose AI material when registering. Treat unprotectable outputs like stock without exclusivity; secure exclusivity via human authorship and/or contractual terms (work-for-hire, warranties, indemnities). Courts have reaffirmed the human authorship requirement (e.g., Thaler case). U.S. Copyright Office+2U.S. Copyright Office+2

Data & biometrics

If your workflows involve face/voice capture (e.g., training virtual models), expect privacy scrutiny—especially where sensitive data or minors are concerned. The FTC has warned about misrepresenting how biometric tech is used and about unsubstantiated AI claims. Map data flows and secure opt-in consent where appropriate. Federal Trade Commission

What to do in the US:

  • Substantiate any performance or “AI-powered” claims; ban fake/AI-generated reviews. Federal Trade Commission

  • Use robust talent releases covering digital doubles and synthetic voice; vet stock and training sets for license scope.

  • For AI visuals that could mislead (e.g., renders not representative of actual product), fix the creative or add clear, unavoidable disclosures.

European Union (EU)

Consumer protection (UCPD) & ad transparency (DSA)

Under the Unfair Commercial Practices Directive (UCPD), visuals must not mislead consumers. That applies to product photos, CGI, and virtual try-ons. The Digital Services Act (DSA) adds ad transparency obligations (clear labeling of ads, who’s behind them, and why you saw them), and bans targeting based on sensitive data and to children; platforms enforce these rules against advertisers. Expect more auditing of creative and targeting justifications. EUR-Lex+2European Commission+2

EU AI Act—synthetic media & transparency

The EU AI Act (now law) includes transparency duties for deployers that create or manipulate content—commonly called “deepfake labeling.” You must disclose that content is AI-generated or manipulated, with narrow exceptions (e.g., law enforcement). Build label/watermark and provenance into your asset pipeline for EU campaigns. EUR-Lex+1

Environmental/green claims in visuals

If AI-generated imagery implies environmental benefits (clean, green iconography, “zero emissions” vibes), remember the EU’s Empowering Consumers Directive amendments to UCPD and forthcoming Green Claims rules: generic, unsubstantiated sustainability visuals/claims are banned or restricted. Align visuals with substantiated claims only. Reuters

What to do in the EU:

  • Bake AI-content labeling into creative ops; coordinate with platform ad libraries and repositories. European Commission

  • Audit for UCPD risks: no exaggerated product performance, scale, or finish that a reasonable consumer would misinterpret.

  • If using environmental iconography, keep a claims matrix tied to evidence.

United Kingdom (UK)

Misleading advertising (CAP/ASA)

The CAP/BCAP Codes (enforced by the ASA) already cover AI visuals: if an ad is misleading, harmful, or socially irresponsible, it will breach the Codes even if labeled as AI. The ASA and CAP have discussed when to disclose AI—there isn’t a blanket legal duty to say “this is AI,” but transparency is expected where omission would mislead. Treat “AI used” as a tool, not a shield. ASA+2ASA+2

Data protection & biometrics (ICO)

The UK ICO has published detailed guidance on biometric recognition and facial recognition; biometric data is sensitive and high-risk, requiring necessity, proportionality, and strong safeguards (DPIAs, minimization, alternatives). The ICO has enforced against unlawful workplace biometric monitoring—expect similar scrutiny if retail operations repurpose face data for advertising or personalization. ICO+1

Competition & platforms (CMA)

The Competition and Markets Authority has set foundation model principles (access, diversity, choice, fair dealing, transparency, accountability). If you’re partnering with large model providers or ad tech, be mindful of obligations that can flow into advertising and consumer protection contexts—especially if your stack relies on a single vendor’s ecosystem. GOV.UK Assets

What to do in the UK:

  • Apply the CAP Code lens to every AI/CGI visual: “Would a consumer be misled if we don’t explain this is CGI?” If yes, either adjust the creative or add a plain-English disclosure. marketinglaw

  • Run DPIAs where faces/voices are processed; prefer non-biometric alternatives when possible. ICO

China (Mainland)

China has taken a front-foot regulatory approach to synthetic media:

  • Provisions on Deep Synthesis and Interim Measures for Generative AI impose labeling/watermarking and other duties on providers and users of AI content, including requirements to prevent misuse and to maintain records.

  • Enforcement and draft measures in 2024–2025 continue to emphasize strengthened labeling and restrictions to curb deception.

If you operate or publish in the PRC, assume you must label AI-generated visuals and comply with content controls and provenance requirements; align with local partners who can implement compliant watermarking and audit logs. China Law Translate+2Carnegie Endowment+2

What to do in China:

  • Integrate visible and invisible labels/watermarks for AI assets.

  • Maintain content review workflows tuned to platform and CAC rules; document prompts, edits, and approvals.

Canada

Deceptive marketing (Competition Act)

The Competition Bureau enforces against false or misleading representations in advertising. AI visuals that overstate product performance (or that imply qualities the product doesn’t have) can trigger action. The Bureau has been active on online advertising transparency and “fake scarcity” cues—expect scrutiny of AI-generated reviews or deceptive product visuals. Competition Bureau+1

Biometrics and privacy (PIPEDA)

The Office of the Privacy Commissioner (OPC) treats biometric data as highly sensitive, and in 2025 issued detailed guidance for businesses processing biometrics under PIPEDA—covering consent, transparency, purpose limitation, and safeguards. If you build synthetic models/avatars from real faces or voices, plan for enhanced consent and security. Privacy Commissioner of Canada+1

What to do in Canada:

  • Treat any face/voice capture as sensitive; conduct a privacy impact assessment and secure meaningful consent. Privacy Commissioner of Canada

  • Avoid deceptive environmental or performance visuals; tie claims to substantiation consistent with the Bureau’s guidance. Competition Bureau

Cross-cutting legal themes for retail AI visuals

  1. Truth in advertising
    The universal baseline: visuals cannot mislead about what, how, or how well a product performs. CGI polish is fine; misrepresentation isn’t. If the medium (3D/AI) creates a non-obvious gap between depiction and the real product (e.g., color accuracy, scale, finish, results), close it with accurate visuals and/or clear, conspicuous disclosures—ideally both. Federal Register+1

  2. Synthetic-media transparency
    The EU AI Act formally obliges disclosure for “deepfakes,” and China imposes labeling/watermarking duties. Even where not strictly required (US/UK), disclose when omission would mislead—and be prepared for platforms to demand it. White & Case+1

  3. Likeness and voice
    Right-of-publicity (US states), image rights (various), and privacy/biometric laws (UK ICO, PIPEDA) converge here. Use explicit consent; cover digital replicas and voice clones in talent agreements; avoid training on personal data without authority. Justia Law+1

  4. Copyright & ownership
    In the US, pure AI output isn’t protected; mixed human/AI works can be, but you must document the human contribution. For global campaigns, assume patchwork outcomes and rely on contracts (license terms, exclusivity, indemnities) rather than copyright alone for AI-heavy assets. U.S. Copyright Office+1

  5. Platform & ecosystem rules
    The DSA requires ad labeling and repositories; platforms will scrutinize your creatives, targeting, and disclosures. Expect knock-on requirements in self-serve ad interfaces (label toggles, prohibited use policies). European Commission

A retailer’s operational playbook (end-to-end)

1) Design and prompt hygiene

  • Maintain prompt libraries with approved phrasing that avoids risky depictions (e.g., banned medical or efficacy claims).

  • Use reference-accurate product CADs and color profiles when generating renders; keep device calibration notes for color-critical categories (cosmetics, apparel).

  • For China/EU, plan synthetic labels/watermarks at export time, not post-hoc. White & Case+1

2) Talent, likeness, and model rights

  • Update talent release templates: include AI training, digital doubles, voice cloning, retraining, derivative uses, geographic scope, term, compensation, and opt-out mechanics.

  • For influencer programs, require disclosures in captions and visuals when material connections exist; monitor and enforce. Federal Register

3) Substantiation & review

  • Build a claims matrix: map each visual claim to evidence (lab data, standards, internal tests).

  • Flag high-risk visuals: before/after, clinical-like depictions, performance animations, environmental claims. Bring legal in early for these. Reuters

4) Privacy & biometrics by design

  • If you process faces/voices (e.g., for virtual try-on avatars), complete a DPIA/PIA, minimize retention, and secure explicit consent where required (UK/Canada). Avoid silent repurposing of footage for training. ICO+1

5) Labeling and provenance

  • Implement dual-layer transparency: on-screen “AI-generated/CGI” tags where material, plus invisible watermarks or C2PA manifests for forensics and platform checks (EU/China expectations). White & Case+1

6) Records & audit

  • Keep versioned assets with prompts, seeds, edit logs, dataset licenses, QA approvals, and territory mappings.

  • Be ready to show “how this was made” to platforms, regulators, or courts within days, not weeks.

7) Vendor governance

  • Contract with model providers and agencies for: IP warranties, no-scrape / no-PII training assurances, watermark support, territory-specific compliance (EU/China), and indemnities.

  • Pressure-test compliance by running red-team reviews on the most sensitive launches.

Jurisdiction-specific checklists

US checklist

  • Claims: Substantiated; no AI review generation; “clear and conspicuous” disclosures for endorsements. Federal Trade Commission+1

  • Likeness: Signed releases; right-of-publicity coverage (including digital replicas, post-mortem where relevant). Justia Law

  • Copyright: Document human authorship; disclose AI portions in registrations. U.S. Copyright Office

EU checklist

  • UCPD: No misleading visuals; environmental depictions tied to substantiated claims. EUR-Lex+1

  • DSA: Ensure platform ad labels/identity are correct; keep a log of audience rationale. European Commission

  • AI Act: Label deepfakes/synthetic media; prep watermark workflows. White & Case

UK checklist

  • CAP/ASA: Disclose AI use when needed to avoid misleading; fix misleading imagery rather than relying on labels. ASA+1

  • ICO: DPIAs for biometric projects; necessity & proportionality; alternatives to biometric capture where possible. ICO

China checklist

  • Label & watermark synthetic media; follow content restrictions and maintain provenance logs. China Law Translate

Canada checklist

Risk hotspots in retail use cases

  1. Beauty & personal care “results”
    AI-smoothed skin, enhanced shine, or de-aged models can imply effects the product can’t deliver. Use controlled conditions statements and “simulated effect” labels where permitted, or switch to real-world visuals.

  2. Apparel & color-critical goods
    AI/CGI color shifts can misrepresent shade/finish. Include color disclaimers, standardize capture and display profiles, and show real photos alongside renders.

  3. Furniture & large items
    3D staging may distort scale. Add dimensional references and AR measurement prompts; avoid camera angles that systematically exaggerate size.

  4. Eco and ethical claims
    AI-generated pastoral imagery can imply sustainability. Tie any “green” visuals to substantiated, specific claims—or remove. Reuters

  5. Virtual try-on & avatars
    If avatars are built from real customers/employees, you’re handling biometrics: do a DPIA/PIA, secure consent, minimize retention, and give opt-outs. ICO+1

Governance artifacts you should have on day one

  • AI Visuals Policy (internal): Defines when AI/CGI may be used; red-flag categories; disclosure standards by channel/territory; approval gates; and recordkeeping requirements.

  • Talent & Influencer Addenda: Expressly cover digital replicas, voice synthesis, training, retraining, derivatives, moral rights waivers (where lawful), and post-term take-downs. Justia Law

  • Claims Substantiation Dossier: Evidence mapped to each recurring claim type; pre-approved copy blocks. Federal Register

  • Biometrics DPIA/PIA Templates: For any project capturing or inferring face/voice features; include alternatives analysis. ICO

  • Jurisdiction Matrix: Per-market labeling rules (EU AI Act/China), ad transparency (DSA), and platform specifics. White & Case+2China Law Translate+2

Sample policy language (you can adapt)

1. Non-misleading principle
“We will not publish AI-generated or CGI visuals that could materially mislead a reasonable consumer about a product’s appearance, size, performance, or environmental attributes. Where AI/CGI is used and omission could mislead, we will add clear, conspicuous disclosures and/or corrective context.”

2. Synthetic-media transparency
“For EU campaigns and where required by law or platform policy, synthetic or manipulated media will include visible labels and, where feasible, invisible watermarks or C2PA manifests. For China, we will apply CAC-compliant labels/watermarks.”

3. Likeness and voice
“We will obtain written consent for any use of a person’s name, image, likeness, or voice, including digital replicas and voice synthesis, and we will honor revocation rights where applicable. Contracts will cover training, retraining, and derivative uses.”

4. Claims substantiation
“All efficacy or performance depictions (including simulations and animations) require documented substantiation. Environmental depictions must align with substantiated, specific claims.”

5. Privacy & biometrics
“Any project capturing or inferring biometric identifiers requires a DPIA/PIA, explicit purpose limitation, data minimization, security controls, and appropriate consent.”

6. Recordkeeping
“We will retain prompts, seeds, edit logs, dataset licenses, approvals, and territory mappings for at least X years; we will be able to produce these within Y business days upon regulator/platform request.”

Quick answers to common executive questions

Q: Do we have to label all AI images in the EU?
No—not all. But the EU AI Act requires disclosure for deepfakes/synthetic content (generated or manipulated) with limited exceptions. When in doubt, and especially where omission might mislead, label. White & Case

Q: In the UK, is “AI used” labeling mandatory?
There’s no blanket legal requirement; the ASA/CAP focus is on not misleading. If not labeling would mislead, disclose; otherwise fix the creative. Taylor Wessing

Q: Can we own the copyright in AI-only images in the US?
No—purely machine-generated works are not protected. Combine AI with meaningful human authorship, document it, and use contractual controls for exclusivity. U.S. Copyright Office+1

Q: What’s the highest-risk misstep you see?
Realistic likeness/voice without consent (right-of-publicity/biometrics) and performance depictions that over-promise (FTC/UCPD). Justia Law+2Federal Register+2

Final takeaways per market

  • US: No AI carve-outs; endorsements and claims need disclosures and substantiation; talent likeness rights are paramount. Federal Register+1

  • EU: UCPD + DSA already bite; AI Act adds synthetic-media transparency—start labeling/watermark planning now. EUR-Lex+2European Commission+2

  • UK: CAP/ASA “don’t mislead” standard; ICO treats biometrics as high risk; run DPIAs and avoid unnecessary biometric processing. ASA+1

  • China: Label/watermark synthetic media and maintain provenance logs; align with CAC requirements. China Law Translate

  • Canada: Competition Act bans misleading ads; OPC’s biometric guidance sets a high bar for face/voice processing consent and safeguards. Competition Bureau+1

Appendix: Key sources (for your legal and compliance teams)

  • US: FTC Endorsement Guides (2023) & press releases on deceptive AI claims; USCO AI copyright guidance; Thaler appellate decision on human authorship. Reuters+3Federal Register+3Federal Trade Commission+3

  • EU: UCPD guidance, DSA ad transparency (including Commission notes and industry guidelines); EU AI Act synthetic-media transparency. White & Case+3EUR-Lex+3European Commission+3

  • UK: ASA/CAP commentary on AI in advertising; ICO biometrics guidance and enforcement against unlawful biometric monitoring. ASA+2ASA+2

  • China: Deep Synthesis Provisions; Interim Measures for Generative AI; updates on labeling. China Law Translate+1

  • Canada: Competition Bureau deceptive marketing guidance; OPC biometric guidance for businesses (2025). Competition Bureau+1

Bottom line

AI and CGI are phenomenal creative accelerants—but they raise the compliance bar. If you (1) never mislead, (2) disclose synthetic media where law or context requires, (3) get consent for likeness/voice, (4) document human authorship and your process, and (5) keep great records, you can unlock AI at scale—safely and globally.

If you’d like, I can adapt this into a one-page internal policy plus a creative checklist for your designers, merchandisers, and media buyers.

Retail Marketing, Jurisdiction, AI EthicsFrancesca Tabor