AI Video Consent Is Not Reception: What the SKF Garbo One-Star Review Actually Measures

The clearance gate passed and the ad still failed

AI video consent and reception are two different gates, and only one of them is legal. SKF's 99-second AI-generated Greta Garbo ad cleared estate permission, dodged archival footage, licensed the voice and carried AI disclosure stamps — and still took a one-star review from The Guardian, with film critic Peter Bradshaw calling the synthetic Garbo a Stepford robo-star delivering a dull, humourless script. Consent is not reception.

The reason this case matters is that the full consent process is publicly documented and the reception verdict is too. Almost every synthetic-likeness debate so far has argued about unconsented replicas and the right-of-publicity exposure that follows. SKF is the control case in the opposite direction: full consent, and it still failed. That makes the ad a clean read on what a compliance-driven approval process cannot see. The specifics matter, because the depth of the process is what makes the outcome instructive. When Swedish bearing maker SKF ran the ad, the company did everything the clearance checklist asks for: it obtained permission from the Garbo estate and family, generated the character from text prompts rather than archival footage, licensed a single strip of Garbo's recorded voice, brought in a human actor to fix pacing, and stamped AI disclosure across its productions.

The specifics matter, because the depth of the process is what makes the outcome instructive. Per Nilsson, SKF's head of communication and brand, said transparency was a principle rather than a regulatory chore, and that the estate and family followed the whole process and were happy with it. Reuters reported that the Garbo double was not built from archival photographs or footage: copyright considerations meant the likeness came from text prompts, with ByteDance's Seedream 5 Pro and Google's Gemini Nano Banana Pro for images and Seedance 2 and 2.5 plus Kling AI for motion. The voice was trained on Garbo's voice from one early film, the only archival material SKF had permission to feed to the model.

Against that record, the failure was not a clearance failure — nobody in the chain disputed authority to make the ad. The finished piece read as mediocre rather than reverent. Bradshaw's objection was tonal, not legal: bland, mediocre undeadness, a Stepford robo-star delivering a dull and humourless script that tried to build brand synergy between frictionless ball bearings and Garbo's own all-important smoothness. The Telegraph's Sherelle Jacobs made the same point from another direction, noting the recreation had magnetism without presence. Escaping the uncanny valley, which the animation does achieve, is not the same as a performance anyone believes. What was absent was not a permission slip but a second opinion from someone with no stake in the production.

Why AI video consent and reception are two different gates

This is why the consent gap framing, however well meant, keeps missing the failure mode. Existing work on this site already maps {{link}}, which is the opposite failure mode from the one this case exposes. That work is largely about the legal gate: does the brand hold rights, does the estate have authority, is a disclosure present. Those are real exposures and they deserve the attention they get. But the Garbo case shows the two gates can be fully decoupled. A project can pass clearance and still fail on meaning, because meaning is judged by audiences who never see the paperwork.

The practical consequence is a second approval pass with a different owner. Clearance asks a legal question: are we permitted. Reception asks an editorial one: does this read as respectful, or as an appropriation dressed in courtesy. The second question is answered, when it is answered at all, by whoever is closest to the project and therefore most invested in it landing. A useful reception gate has at least one reviewer who did not build the asset and has no delivery date attached, and the test is whether that reviewer can name the specific moment the piece turns unpleasant — the joke that lands wrong, the synthetic smile that reads as smug, the script line that treats a life as product placement.

Survey work on {{link}} shows how diffuse that resistance is in aggregate, which is exactly what makes it hard to act on. The Garbo case is one instance of that gap with a specific, addressable complaint attached to it.

Existing work on this site already maps the legal exposure that follows unconsented synthetic replicas, which is the opposite failure mode from the one this case exposes.

Survey work on how diffuse the resistance is in survey data shows how diffuse that resistance is in aggregate, which is exactly what makes it hard to act on.

The statutory definition quietly exempts the hardest case

Regulatory design makes the gap sharper rather than closing it. New York's synthetic performer disclosure law, signed December 11, 2025 and effective June 9, 2026, is the first-in-the-nation statute of its kind and amends General Business Law § 396-b to require conspicuous disclosure in advertisements containing a synthetic performer, with civil penalties of $1,000 for a first violation and $5,000 for subsequent ones. The operative definition is narrower than the marketing conversation. A synthetic performer is a digitally created asset intended to create the impression of a human performer who is not recognizable as any identifiable natural performer.

Read that definition carefully and a recognizable digital resurrection of a named, licensed individual sits outside the disclosure trigger — the statute reaches synthetic people, and the estates and celebrities protected by right-of-publicity law are handled by those existing laws instead. That is a defensible division of labour on paper. In practice it means the case most likely to be read as an insult is the case the state disclosure regime was not written to reach. SKF's Garbo ad, the best-documented consent case of the year, would have sat in exactly that gap. This is the same pattern visible across {{link}}: frameworks are converging on the presence of a label rather than the quality of the reading a label makes possible.

None of which is an argument against disclosure. Its practical guidance — inventory campaigns using digital humans, flag synthetic performers in creative review checklists, and put disclosure responsibility into agency and vendor contracts — is the right shape for compliance. The problem is that a compliance checklist keyed to a definition is not a reception strategy. A project can satisfy every element and still produce 99 seconds of humourless undeadness that made a national critic reach for one star. The same shape shows in newsroom metadata, where the IPTC Digital NewsCodes vocabulary records trainedAlgorithmicMedia, algorithmicMedia and composite as properties of the file: provenance is captured precisely, and reception is not a field at all.

This is the same pattern visible across how disclosure regimes are converging worldwide: frameworks are converging on the presence of a label rather than the quality of the reading a label makes possible.

Living talent is now pricing the same risk in advance

The other live signal is that performers are responding to the same asymmetry from the other side of the camera. At the Venice Film Festival in September 2026, George Clooney said he would like to avoid appearing in a tampon commercial twenty years after his death. Usually filed under celebrity anxiety, read alongside Garbo it is closer to a repricing. A digital double costs a fraction of live talent, and any performer who can foresee that asymmetry has an incentive to demand synthetic-use terms while they can still negotiate them.

Bradshaw's closing line — that guardrails were never more badly needed — is the press's version of the same argument. The estate and the critic agreed on the facts, and neither was the problem. The industry does not have a permissions problem in the Garbo case; it has an interpretation problem, and interpretation is the one thing a permissions workflow was never designed to produce. The answer is not to stop: generative resurrection is genuinely capable now in ways hand-keyed rotoscoping and stand-in body doubles never were, and the collapse in cost is exactly why it will keep happening. The answer is that clearance and reception become two owned decisions with two different owners, and that a project which cleared every legal gate still owes someone an honest read on whether the thing it made is any good.

What to change on Monday

The first change is procedural: name a reception reviewer outside the creative team, with authority to stop a launch that cleared legal review. The second is a pre-mortem against the specific's own intention. Garbo retired from public life in 1941 and became almost mythically private, and the ad's best moment is a joke built on that withdrawal. Ask whether the line reads as an affectionate reference the subject would recognise, or as a brand using someone's restraint as an asset. If the team cannot answer confidently, the joke is the problem, not the clearance.

The third is to stop treating a disclosure mark as the finish line. A label satisfies a statutory definition; it does not make a synthetic performance land. Measure reception on its own terms — run a small unreleased cut past reviewers who did not approve it, and track whether the criticism lands on craft or on consent. In the Garbo case it landed squarely on craft: dull, humourless, a parody. That is a criticism production can act on, and it arrives whether or not a legal review signed off first.

Teams that already gate the idea before production can reuse {{link}} here, because the question of whether a synthetic performer was load-bearing for the campaign idea is identical in structure and simply applied one stage earlier. Consent and necessity are both upstream questions. Reception is the one that only exists once there is something concrete to react to, which is why it needs its own owner rather than a line in someone else's checklist.

Teams that already gate the idea before production can reuse the same upstream reasoning here, because the question of whether a synthetic performer was load-bearing for the campaign idea is identical in structure and simply applied one stage earlier.