A court is dealing with what happens when somebody else's face becomes the source material
A South Australian court has heard a victim impact statement in a prosecution involving non-consensual explicit deepfake imagery. William Hamish Yeates pleaded guilty to four charges, including creating or altering sexual material without consent and using a carriage service in a harassing or offensive way. Reporting describes the material as having been created by attaching the victim’s face to explicit imagery and says this is the first prosecution of its kind under newer federal and state legislation. Sentencing is scheduled for next month.
Why visual creators should care
The source photograph does not have to be stolen or commercially valuable for serious rights issues to arise. A person's identifiable face can itself become the source asset for an entirely different synthetic depiction.
Practical takeaway
For professional likeness work, do not treat possession of a portrait as permission to generate new representations of the person. Record synthetic-transformation permission separately from permission to photograph, edit or publish the original.
What to watch next
Watch the sentencing decision and how Australian courts apply the newer offences to the act of creating, altering and distributing synthetic visual material.
Sources
An Arkansas case puts the photographer's own client files at the centre of an AI deepfake prosecution
A Bentonville photographer accused of using professional photographs of juvenile clients as source material for AI-generated sexual imagery returned to court on September 21 for an omnibus hearing. Russell Edward Bloodworth has pleaded not guilty. Reporting says he faces more than 200 charges, including 100 counts relating to unlawful creation or distribution of deepfake visual material. Investigators allege thousands of generated images were found and that professional client photographs were among the source material.
Why visual creators should care
The allegation presents an extreme version of a broader professional issue: photographers can possess high-resolution likeness files because a client trusted them for one specific purpose. Technical possession of those files does not create unlimited rights to generate derivatives from them.
Practical takeaway
Studios handling identifiable client photographs should restrict access to originals and treat AI generation from client files as a separate authorised use. A model release for the original assignment should not silently become permission for unrelated synthetic derivatives.
What to watch next
Watch the pretrial proceedings and how prosecutors establish the relationship between original client photographs and allegedly generated derivatives.
Sources
An AI avatar platform is treating the face and the photograph as two different kinds of data
Grovly updated its Biometric Data Policy on September 21. When a user uploads a real photograph to create a talking avatar, Grovly says an automated face-detection service analyses facial attributes and the photograph is sent with generated audio to Hedra for animation. Voice cloning uses Cartesia to derive a voice model. Grovly classifies facial geometry derived from an identifiable photograph and a cloned voiceprint as biometric data. It requires a separate affirmative consent step before processing rather than relying on general acceptance of its Terms, records when consent was given and provides deletion and withdrawal mechanisms. It says biometric data is not used for advertising or general model training.
Why visual creators should care
A portrait entering an AI workflow can become more than an image file. The system may extract or derive information from the person represented inside it, creating a separate biometric-data layer alongside ordinary copyright and image rights.
Practical takeaway
When commissioning synthetic presenters or avatars, record the source photograph, likeness permission, biometric processing consent, processor, retention period and deletion procedure separately.
What to watch next
Watch whether avatar platforms begin supplying portable consent receipts that can travel with the resulting synthetic presenter or digital replica.
Sources
A virtual-fashion service is documenting exactly when a photograph leaves the app
Unmarble updated its privacy policy on September 21 for its AI outfit-visualisation service. The company says only the particular photographs selected for a requested design are sent to Google Gemini through the Google Gen AI API. Before the first generation, the user receives a separate consent screen and no image is transferred if consent is declined. Unmarble says personal photographs are not used to train its own models and that Google is contractually restricted from using the submitted images for its own purposes or model training. Uploaded and generated images remain in the user's private gallery until deleted.
Why visual creators should care
Reference-image workflows often hide the handoff to another AI company. Here, the platform identifies the downstream model provider, which images leave the service and what the provider is permitted to do with them.
Practical takeaway
For client and fashion imagery, record the actual processor behind the front-end service. A privacy review should identify not only the app receiving the photograph but every AI provider that subsequently receives it.
What to watch next
Watch whether professional visual platforms make processor-level disclosure and pre-transfer consent standard for identifiable photographs.
Sources
A face-swap service makes the cloud copy of the photograph part of the privacy record
AI Swap Face updated its privacy policy on September 21. Despite the browser-facing workflow, photographs and videos selected for a face swap are sent to cloud GPUs for processing rather than being transformed locally on the device. Submitted source files and generated swaps are stored on the service's cloud servers so users can access previous results. The company says these images are used only to provide the face-swap service and are not sold, repackaged or used for AI model training; individual swaps and associated source files can be deleted.
Why visual creators should care
A simple visual interface can make an AI edit appear local even when identifiable photographs are travelling to remote infrastructure and being retained. For client imagery, that distinction can affect confidentiality and privacy obligations.
Practical takeaway
Before putting a client portrait into an AI editor, check processing location, storage duration and deletion controls. Browser-based does not necessarily mean on-device.
What to watch next
Watch whether image-editing services begin offering genuinely local face-processing modes for professional and privacy-sensitive work.
Sources
A design competition is drawing the AI boundary at the source artwork
The 2026 Slowdown Art Competition opens September 22 for artists worldwide. Its rules require submitted artwork to be entirely original and human-made and explicitly prohibit AI prompts and AI-generated elements. Six winning designs will be produced as woven blankets. Artists keep ownership of their work, while selected winners grant Slowdown defined exclusive commercial rights to produce and promote the resulting product for up to three years, plus a six-month sell-through period.
Why visual creators should care
The competition separates two questions that are sometimes blurred together: whether AI-generated material is eligible at all, and what commercial rights an artist grants after selection. The human-made requirement does not require the artist to surrender copyright.
Practical takeaway
For competitions and commissions, keep eligibility and licensing separate. Verify whether AI-assisted or generated elements are permitted, then separately check what rights the organiser receives in work that qualifies.
What to watch next
Watch whether competitions increasingly define 'human-made' by process rather than relying on visual inspection or unreliable AI detectors.
Sources
