An AI clone is now competing with the influencer whose face made the endorsement valuable
Singapore influencer Dewy Choo discovered an unauthorized AI-generated version of her face and voice being used in a TikTok advertisement for false eyelashes. The commercial connection is particularly striking because Choo had previously promoted the same brand herself. The Straits Times also found AI advertisements reproducing Chinese actor Zhang Linghe, including distinctive facial details, to sell products to Singapore users. Choo reported the synthetic advertisement and it was removed.
Why visual creators should care
Synthetic likeness is no longer only a deepfake or reputation problem. It can compete directly with a creator’s paid endorsement work. An advertiser can potentially reproduce the identity and recognisability that a photographer, performer or influencer built commercially without paying the person who supplied that value.
Practical takeaway
For endorsement and talent agreements, address synthetic reuse explicitly. Permission to photograph or film somebody for one campaign should not silently become permission to generate new performances, voices, poses or advertisements later.
What to watch next
Singapore’s Online Safety Commission became operational in June and is expected to address inauthentic-material abuse as its remit develops. Watch how quickly victims of commercial deepfakes gain practical takedown or account-restriction remedies.
Sources
Film professionals are being paid to turn their own production knowledge into AI training material
The Guardian reports that experienced Hollywood writers, directors and producers are taking temporary work training AI systems in skills including screenwriting, production planning and pitch creation. Reported rates vary widely, from about $12 to $200 an hour. Some participants describe the work as economically necessary during a prolonged production slowdown while also worrying that they are helping automate parts of their own profession.
Why visual creators should care
Training data is not always a scraped archive. Increasingly it can be professional expertise deliberately supplied by working creators. That raises different contractual questions: what examples are being contributed, how permanently may the company use them, whether confidential industry knowledge enters the system and whether any downstream participation or compensation remains.
Practical takeaway
If you accept AI-training work, read the rights clause as carefully as you would a commission. Check who owns the examples you create, whether the company can retain and reuse them indefinitely, whether confidential client knowledge is excluded and whether your name or professional reputation can be associated with the resulting model.
What to watch next
Watch whether guilds and creative organisations develop standard terms for paid AI-training work and whether contributors begin demanding restrictions on reuse, attribution or downstream commercial exploitation.
Sources
Canva’s standard AI terms and beta terms do not create the same data bargain
Canva’s AI Product Terms, effective June 26, say users retain rights in their inputs and, as between Canva and the user, own outputs to the maximum extent permitted by law. Canva cautions that outputs may not be unique and that an output incorporating Canva library material remains subject to the underlying content licence. Separate Canva AI beta terms create a more expansive data arrangement: beta inputs, outputs and interaction data may be used to develop, train, test, evaluate and fine-tune machine-learning systems, and the terms state that ordinary account AI-consent settings do not apply to that beta data.
Why visual creators should care
A designer can move from a normal AI feature to an experimental version of apparently the same service and cross into materially different rules for client files. Product status therefore becomes part of the rights record, particularly where unreleased designs, photographs or brand assets are uploaded.
Practical takeaway
Check whether the feature is standard, preview or beta before uploading client material. Do not assume an account-level AI-training preference overrides feature-specific beta terms.
What to watch next
Watch what happens when beta features graduate into Canva’s standard product and whether training permissions collected during the beta period remain attached to previously uploaded material.
Sources
Pinterest says AI provenance is part of the output and creators should not strip it away
Pinterest’s current business terms for its Creative GenAI features say users must not misrepresent AI-generated output as human-generated and must not alter or remove watermarks or content-authenticating metadata supplied with the output. Pinterest may also add disclosure language where legally required or appropriate. Its separate AI-label system uses both image metadata and classifiers and provides an appeals route where creators believe a Pin has been incorrectly labelled.
Why visual creators should care
For designers and advertisers, provenance is becoming a contractual obligation as well as a technical feature. Removing a label or metadata field because it interferes with a layout can therefore create a terms problem even before any statutory AI-disclosure requirement is considered.
Practical takeaway
Before stripping metadata from generated advertising assets, check the platform terms that produced them. Keep a clean visual master if needed, but preserve the provenance-bearing original and any required disclosure in the published version.
What to watch next
Watch how Pinterest’s AI labels interact with the EU AI Act and whether advertisers receive stronger in-product warnings when an export or downstream workflow removes required provenance information.
Sources
A 3D artist’s lawsuit asks whether an open online asset is also open AI training data
Los Angeles 3D artist and visual-effects creator Austin Beaulier filed proposed class actions against Meta, Nvidia, Roblox and Microsoft alleging that his work and millions of other 3D models were incorporated into datasets used for generative-AI systems. The complaints focus in part on material distributed through repositories under Creative Commons licences containing conditions such as attribution or restrictions on commercial use. The defendants had not been found liable; these remain allegations being litigated.
Why visual creators should care
3D artists sit in an especially exposed part of the visual-training ecosystem because large public repositories make high-quality structured assets easy to collect. A licence allowing someone to download and use a model does not necessarily answer whether they may remove attribution, build a commercial training dataset from it or train a generator that produces new assets.
Practical takeaway
When publishing 3D models or other reusable visual assets, keep the licence version and publication record. If you depend on Creative Commons conditions, preserve evidence showing exactly which conditions accompanied the file when it was made available.
What to watch next
Watch how the courts treat licence conditions, copyright-management information and the distinction between accessing publicly available models and using them as commercial AI-training material.
Sources
One film festival is drawing the disclosure line at authorship and performance rather than software
The Okanagan International Film Festival’s current rules require filmmakers to disclose AI where it generated significant narrative elements, primary visual or audio performances, or replicated identifiable people. By contrast, AI-assisted rotoscoping, upscaling, noise reduction, VFX assistance, colour work and editing do not require disclosure unless the technology materially affects authorship or performance. The rules also require compliance with copyright, consent, likeness and applicable guild requirements.
Why visual creators should care
This is closer to the distinction many working filmmakers need than a blanket 'AI used' checkbox. Software increasingly contains machine-learning features by default, while generating an actor or a substantial part of a story changes the nature of the creative contribution much more significantly.
Practical takeaway
Build disclosure around material creative intervention. Record whether AI generated a performance, person, scene or narrative element separately from routine machine-learning tools used in post-production.
What to watch next
Watch whether film festivals and clients converge around this kind of materiality test and whether they begin asking for production records where a synthetic performer or replicated likeness appears.
Sources
