A disclosed political avatar is testing the limits of transparency
An AI-generated avatar of Jair Bolsonaro appeared at the launch of his son Flavio Bolsonaro’s presidential campaign in Brazil. The video clearly disclosed that it was made with AI. Political opponents have nevertheless challenged it in electoral court, arguing that the synthetic appearance may breach electoral restrictions and allow Bolsonaro to communicate indirectly while he remains restricted from public communication under house arrest. Flavio Bolsonaro’s legal team argues that the disclosed avatar was not deceptive.
Why visual creators should care
The case demonstrates that transparency answers only how the media was produced. It does not answer whether the represented person authorised it, whether the avatar may speak on that person’s behalf or whether the use circumvents a restriction affecting the real individual.
Practical takeaway
For any recognisable synthetic spokesperson, document the person’s permission, who approved the script, who has authority to speak on their behalf and whether any legal or contractual restrictions apply to the real person.
What to watch next
Watch whether Brazil’s electoral court distinguishes deceptive deepfakes from disclosed synthetic surrogates and whether an avatar may lawfully communicate a message the real person is prohibited from delivering.
Sources
One synthetic campaign video can misuse both the person and the photograph
A political campaign video in Bridgeport, Connecticut appears to use AI-generated versions of Mayor Joe Ganim and Probate Judge Paul Ganim, including fabricated dialogue and synthetic voices. CT Insider reports that at least one still photograph used to construct the video appears to have come from its archive without permission. The campaign did not deny using AI but defended the political message. The video reportedly lacked a clear synthetic-media or satire disclaimer.
Why visual creators should care
The people depicted may have likeness, reputation or political-media claims, while the photographer or publisher has a separate copyright interest in the source photograph. Permission involving one right does not clear the other.
Practical takeaway
Before turning a still photograph into generated video, clear both the source photograph and the depicted person. A licence to publish the still does not automatically permit animation, voice synthesis or fabricated performance.
What to watch next
Watch whether the source photograph leads to a copyright claim and whether Connecticut revives proposed rules requiring persistent disclosure on synthetic political advertising.
Sources
Adobe’s updated retirement list shows how quickly a professional AI workflow can disappear
Adobe updated its generative-model retirement register on July 28. The list confirms that OpenAI Sora 2, Runway Aleph and Runway Gen-3 Alpha Turbo were deprecated in Adobe products on July 23. Earlier removals include Google Imagen 3 and 4, Veo 2 and Veo 3, Moonvalley’s Marey, Ideogram and Pika 2.2. Adobe says partner models may be removed because of low use, quality concerns, expiring contracts or the provider ending support.
Why visual creators should care
A creator may be able to produce an approved campaign asset today but be unable to reproduce or revise it later because the model behind the Adobe interface has disappeared. Recording only that the work was made in Firefly or Creative Cloud is therefore insufficient.
Practical takeaway
Record the host application, external model provider, exact model name, date, settings and important prompts. Save approved generations and intermediate assets locally before a model is deprecated.
What to watch next
Watch whether Adobe introduces project warnings, migration support or persistent records showing which retired partner model generated an existing asset.
Sources
Runway’s Gen-3 Alpha Turbo disappears on July 30
Runway says Gen-3 Alpha Turbo will no longer be available after July 30. The older Gen-3 Alpha model was retired on July 8. Runway directs users towards Gen-4.5 for text-to-video and image-to-video work, Animate Frames for keyframes and Edit Studio Aleph 2.0 for video-to-video editing.
Why visual creators should care
Switching to a replacement model is not a neutral software update. The same source footage and instructions can produce a visually different result, affecting continuity, approved character appearance and the ability to recreate client work.
Practical takeaway
Complete essential Gen-3 revisions before retirement and export anything likely to be needed later. Record which shots depend on Gen-3 Alpha Turbo and test replacement models before promising consistent future revisions.
What to watch next
Watch whether previously generated projects remain editable after retirement and whether Runway preserves the model identity and settings in project histories.
Sources
FLUX 3 turns a reference image into a potentially reusable character, voice and moving performance
Black Forest Labs has opened early access to FLUX 3, a multimodal foundation model trained across images, video and audio. It can produce video with native sound, animate a starting image, use images as visual references, transfer a central character from a source video into a new scene, generate multilingual dialogue and chain clips into longer sequences. FLUX 3 Image is scheduled to enter early access separately.
Why visual creators should care
A source image or clip is no longer simply influencing the appearance of one output. It can become the basis for a consistent character moving, speaking and appearing across new contexts. That expands the clearance question from image use to synthetic performance and future reuse.
Practical takeaway
Before supplying a person, character, photograph or video as a FLUX reference, confirm that the licence covers animation, new scenes, generated dialogue and repeated character use. Keep a register connecting every generated character to its source assets and permissions.
What to watch next
Watch the final FLUX 3 terms, safety controls, Content Credentials or watermarking arrangements and whether early-access users receive clear restrictions on face, voice and character references.
Sources
TikTok says synthetic spam is beginning to displace original creators
TikTok is testing improved systems aimed at detecting accounts dedicated to mass-producing AI-generated spam, particularly around politics, current affairs, finance and health. The platform says such content can crowd out authentic creators. TikTok has also joined the C2PA Steering Committee and says it has labelled more than three billion videos using Content Credentials, creator disclosures and invisible watermarking.
Why visual creators should care
Platform policy is beginning to assess the behaviour of an account and the industrial scale of production, rather than looking only at whether each synthetic post carries a label.
Practical takeaway
Keep dated master files and publication records. Creators using AI deliberately should also retain process evidence that distinguishes an authored body of work from repetitive automated output.
What to watch next
Watch how TikTok defines AI spam, whether legitimate AI-assisted creators are falsely restricted and what appeal process exists when enforcement affects reach or monetisation.
Sources
Meta has acknowledged that public visibility is not permission for generative likeness use
Meta launched Muse Image with a feature that allowed users to reference public Instagram accounts through an @-mention when generating new images. Meta removed the feature three days later, saying that feedback showed it had missed the mark. Muse Image continues to support image generation, editing and blending of uploaded photographs.
Why visual creators should care
A publicly accessible portrait or profile does not automatically become an authorised generative reference. Meta’s withdrawal provides an unusually clear platform acknowledgement of that distinction.
Practical takeaway
Obtain specific consent before using a public profile, portfolio portrait or social photograph to construct a recognisable generated person, especially for advertising or commercial publication.
What to watch next
Watch whether Meta introduces a true opt-in likeness system before Muse Image capabilities enter its advertising products.
Sources
A creator’s public archive can now be turned into a fraudulent commercial performance
Creators and public figures report fraudulent advertisements using AI-generated versions of their faces and voices to promote products they never endorsed. Existing photographs and social videos are being transformed into new synthetic commercial performances, while victims describe difficulty identifying advertisers and securing permanent removal.
Why visual creators should care
For creators whose professional identity depends on public visibility, the material used to build a reputation can also become the raw material for an unauthorised endorsement.
Practical takeaway
Maintain an archive of genuine campaigns and authorised endorsements. Preserve the complete fraudulent advertisement, account, destination page, publication date, platform information and the likely source photograph or video.
What to watch next
Watch whether platforms create dedicated notice systems for synthetic commercial impersonation and whether repeat uploads can be blocked using reference files supplied by the affected creator.
Sources
Courts are still deciding AI disputes through familiar rights
Researchers reviewed 559 US federal court opinions in which AI played a role in the parties’ arguments. They found that courts generally rely on established doctrines rather than developing one unified body of AI law. Relevant disputes continue to be decided through claims involving copyright, contract, privacy, fraud, evidence and other conventional legal rights.
Why visual creators should care
Describing something as an AI dispute does not identify the actual legal wrong. A visual creator still needs evidence showing ownership, consent, licence scope, false representation or misuse of identity.
Practical takeaway
Structure project records around individual rights. Preserve copyright ownership, model and performer consent, client permissions, publication evidence, platform correspondence and proof of financial or reputational harm.
What to watch next
Watch whether synthetic-likeness, visual-training and generated-evidence cases begin producing more consistent rules or remain dependent on the traditional claim available in each dispute.
Sources
