Alamy has put its ban on AI-generated imagery directly into the contributor contract
Alamy’s current contributor contract says submitted photographs, video, illustrations and other imagery must not be created using artificial-intelligence image-generation tools. Submission of AI-generated content is expressly treated as a material breach. The contract also contains a separate rule for news imagery: editorial submissions must be free from manipulation, alteration or post-production, including AI-powered editing, where the change alters truth, context or editorial integrity.
Why visual creators should care
For stock and editorial photographers, the distinction is important. AI generation can make an asset contractually ineligible for the library, while AI-assisted processing of a genuine photograph becomes a separate question centred on whether the edit changes what the photograph claims happened.
Practical takeaway
If you distribute the same photography through several agencies, maintain separate delivery versions where their AI-editing rules differ. For live news, preserve the original capture and keep a clear edit history showing that any AI-assisted operation did not change factual content.
What to watch next
Watch how Alamy interprets generative object removal, expansion and other hybrid edits and whether its quality-control process develops clearer distinctions between generative editing and assistive tools such as denoise.
Sources
Pexels lets contributors switch off the use of their photographs for AI training
Pexels updated its AI and ML FAQ on September 4. It says contributor content can be used for several machine-learning purposes, including improving search, building systems that detect AI-generated material and supporting generative-AI projects across Canva products. Contributors who do not want their imagery used for those AI and machine-learning initiatives can opt out through their profile settings and have the content removed from training use.
Why visual creators should care
A photograph can remain ordinary human-created stock while still becoming part of an AI development pipeline. For photographers, that makes permission for AI training a separate question from permission for people to download and reuse the photograph under the normal Pexels licence.
Practical takeaway
If you contribute to Pexels, check the AI and ML setting rather than assuming the ordinary image licence answers the training question. Where a no-training position matters to you or a client, save a dated record showing that the opt-out was active.
What to watch next
Watch whether Pexels provides contributor-level records of when the opt-out took effect and how it handles data already incorporated into completed model-training runs.
Sources
Midjourney’s updated privacy policy shows how much visual data can sit around a generation
Midjourney’s Privacy Policy was updated September 3. It says personal data handled through the service can include text and image prompts, photographs, videos, documents and messages supplied by users. The policy also covers data collected through the process of training Midjourney machine-learning algorithms and information obtained from third-party sources including public databases, commercial sources and the public internet.
Why visual creators should care
For professional creators, the input is often more sensitive than the finished output. Reference photography, unpublished campaign material or client assets can carry personal or commercial information independently of whatever image the model eventually generates.
Practical takeaway
Treat reference images and other uploads as data entering the service, not merely temporary creative instructions. For confidential work, check both the platform’s privacy terms and the visibility setting before uploading client material.
What to watch next
Watch whether Midjourney provides more granular enterprise controls around uploaded reference material, retention and training, particularly as image editing becomes a larger part of the service.
Sources
European agencies now have a clearer test for when an AI image actually needs a visible label
VIA Nederland has published guidance for creative agencies applying Article 50 of the EU AI Act. For image and video work, it says the central question is whether generated or manipulated content can mislead viewers into thinking a depicted person, object, place or event is authentic. Realistic synthetic people, face replacement and similar manipulations can trigger disclosure, while ordinary technical adjustments to colour or lighting do not automatically do so. Where disclosure is required, invisible metadata alone is not enough: the viewer must be able to perceive the notice.
Why visual creators should care
This gives agencies a more useful production test than simply asking whether Midjourney, Firefly or another AI tool appeared somewhere in the workflow. Material alteration of reality and ordinary AI-assisted post-production are not being treated as equivalent.
Practical takeaway
Add an authenticity check to campaign approval. Ask whether a reasonable viewer could mistake the generated or altered visual for a real person, event or environment. If the answer is yes, plan the disclosure into the creative rather than relying only on C2PA or other embedded metadata.
What to watch next
Watch how the European Commission and national regulators apply the deepfake test to product advertising, synthetic models and hybrid shoots combining real actors with generated environments.
Sources
Danny Boyle’s new film uses AI to bring old photographs to life
Director Danny Boyle says his forthcoming Netflix film Ink contains roughly 30 seconds of AI-generated imagery. The production used AI to animate a small number of old photographs and to generate mice appearing in the newsroom setting. Boyle described AI as another tool within the visual-effects toolbox but said it should be used respectfully and that actors should be paid when an AI embodiment replaces or reproduces them.
Why visual creators should care
The archival-photography use is a useful example of where generative filmmaking is moving. A historical photograph is no longer necessarily used only as a static record; filmmakers can now manufacture movement and new visual information around it.
Practical takeaway
If archival photography is being animated or extended with AI, distinguish the licensed original photograph from the newly generated performance or movement. Check whether the source-image licence permits derivative manipulation and whether recognisable people create additional likeness issues.
What to watch next
Watch how Netflix presents or discloses the synthetic sequences and whether credits or production documentation identify the tools and source photographs used.
Sources
Pexels wants human-made photography in the library even while human photography can support its AI work
Pexels updated its contributor guidance on September 5 to reiterate that photographs and videos created using generative AI cannot be uploaded to the platform. That sits alongside its separate September 4 AI and ML policy, under which human-created contributor content can support authenticity detection, search and generative-AI initiatives unless the contributor opts out.
Why visual creators should care
The two policies create an important distinction for photographers. Pexels is protecting the downloadable library as a collection of human-created visual work while still treating that library as potentially useful data for building AI systems.
Practical takeaway
Do not assume that a platform’s refusal to accept generated imagery means your genuine photography is excluded from AI use. Upload eligibility and training permission are separate policies and need to be checked separately.
What to watch next
Watch whether Pexels eventually exposes AI-training status at individual-image level and whether its ban expands to photographs that contain limited generative edits.
Sources
