The photographer-versus-LAION case may be turning into a European copyright question
Germany’s Federal Court of Justice heard photographer Robert Kneschke’s case against LAION on September 3. LAION assembled a dataset containing about 5.85 billion image-text pairs that can be used for AI development. The dataset stores links rather than the source photographs themselves, but LAION downloaded images during the process of analysing whether photographs matched their associated descriptions.
Lower German courts found that use covered by text-and-data-mining exceptions. During the Federal Court hearing, the copyright panel indicated that it sees unresolved questions of European law and may refer those questions to the Court of Justice of the European Union. No final decision has yet been issued.
Why visual creators should care
A CJEU referral could move the dispute beyond one photographer and one dataset. It could influence how Europe interprets the relationship between copyright, research-related text and data mining, commercial AI development and rights holders’ ability to reserve photographs against machine use.
Practical takeaway
Continue treating AI-training reservations as evidence worth preserving. Keep copies of the publication terms, robots or NoAI signals and stock-library restrictions that applied when photographs were made available online. The legal value of those signals remains contested, but reconstructing them years later is much harder.
What to watch next
Watch for the BGH’s formal decision on whether it will refer questions to the CJEU and, if it does, exactly how those questions distinguish dataset construction from the later training of a generative model.
Sources
A new international cinema declaration says AI can assist creation but human-created works remain the foundation
The Lumière Declaration was adopted on September 7 at the first international summit dedicated to cinema and the moving image. Its signatories say artificial intelligence can strengthen creativity and innovation provided it acts as a driver of creation, while stating that human-created works remain the foundation of the creative industries and that human creation should stay at the heart of works. The declaration also calls respect for intellectual property foundational to the moving-image industry. France and South Korea separately launched the Lumière Partnership with €1 billion of announced investment, intended to support audiovisual businesses while promoting human creation, copyright and cultural diversity.
Why visual creators should care
This is not binding copyright law, but it is a significant policy statement from governments, film-industry leaders, institutions and creators about the principles they want attached to AI adoption in professional screen production.
Practical takeaway
For film and agency work, record AI as part of a production chain rather than an invisible substitute for human authorship. Where generative tools affect performers, source works or substantial creative contributions, document permission, credit and value-sharing explicitly.
What to watch next
Watch whether the Lumière principles become concrete procurement, funding or production conditions and whether the €1 billion partnership gives preference to projects following particular creator-rights or AI-transparency standards.
Sources
X starts paying around a new definition of original visual content today
X begins rolling out applications for its Original Content Rewards programme on September 8, replacing the former Creator Revenue Sharing system. X defines eligible original content as work personally created, written, filmed, designed or produced by the account holder. The policy expressly includes original photographs and video, alongside graphics, reporting and commentary. Straight copying, downloading another creator’s material and re-uploading it, or reposting without meaningful transformation does not qualify. The published rules do not clearly establish whether a fully AI-generated image is automatically eligible or ineligible simply because AI created it.
Why visual creators should care
Platform monetisation is starting to depend on a judgement about visual originality. Photographers and filmmakers are explicitly within the rewarded category, while accounts built around copying and re-uploading other people’s visual material lose eligibility.
Practical takeaway
If monetising original photography or video on X, keep authorship evidence and publish from the account that actually created the work. Do not assume the platform’s definition of 'original' is identical to copyright authorship or that an AI-generated image will necessarily be treated the same way as an original photograph.
What to watch next
Watch how X classifies AI-generated and heavily AI-assisted visual posts under the new programme and what evidence creators can provide if genuine original photography is wrongly treated as reused material.
Sources
DeviantArt treats 'made with AI' and 'do not train on this' as two completely different labels
DeviantArt’s current submission system has separate signals for work created with AI tools and work that is not authorised for inclusion in third-party AI-training datasets. The Created with AI tools label is automatically added to DreamUp work and can also be triggered by metadata that clearly requires the designation. The separate NoAI label tells third-party dataset builders that the artwork is not authorised for AI-model training and can be applied individually or in bulk across a gallery.
Why visual creators should care
The two labels answer different questions. One describes how an image was produced. The other expresses what somebody else is permitted to do with it. Treating them as interchangeable loses an important part of the creator’s rights position.
Practical takeaway
For portfolio work, keep AI-production disclosure and AI-training permission as separate fields in your own records. A conventional human-made artwork may still need a NoAI reservation, while an AI-assisted work may need disclosure without implying anything about downstream training permission.
What to watch next
Watch whether NoAI-style reservations converge into a broader interoperable standard and whether dataset operators begin providing auditable evidence that machine-readable creator reservations were respected.
Sources
A new AI-image contract limits the platform’s permission over your photograph to performing the task you asked for
ImageHub’s Terms of Service took effect September 7. Users retain all rights they already have in photographs and prompts uploaded to the service. The licence granted to ImageHub is described narrowly: it can store, transmit and process the material, including passing it to underlying AI service providers, only for the purpose of carrying out the requested task. Generated results may be used commercially, but ImageHub does not promise exclusivity, copyright protection or freedom from third-party rights. Users must also possess the rights associated with people appearing in uploaded images.
Why visual creators should care
AI image aggregators can sit between the photographer and several underlying model providers. A narrowly stated processing licence is useful, but it still means the source photograph may leave the host platform and enter another provider’s infrastructure.
Practical takeaway
For confidential or client photography, record both the host platform and any underlying AI provider where that information is available. A promise that your upload remains yours does not mean the file never leaves the platform’s own systems.
What to watch next
Watch whether image aggregators begin naming the exact provider that handled each edit and providing model-level retention and no-training guarantees.
Sources
Real-person consent is moving directly into the rules for face swaps and synthetic contact
Luxoret’s current content policy applies across image, video and other generative tools. For face swaps, character swaps, clothing changes, kissing and hugging video and other real-person transformations, users may upload a real person only if it is themselves or they have that person’s explicit permission. The platform separately prohibits generating realistic events as though they were genuine recordings and requires users to disclose AI generation when material published outside Luxoret could reasonably be mistaken for a real photograph or recording. Provenance metadata is attached where the underlying model supports it.
Why visual creators should care
A general licence to possess or publish someone’s photograph does not necessarily authorise manufacturing a new action, outfit, embrace or performance involving that person. Tool-specific consent requirements are beginning to recognise that distinction.
Practical takeaway
For synthetic people, make the release describe the transformation you actually intend to perform. Permission to photograph a model should not automatically be treated as permission for face swaps, synthetic intimacy, clothing replacement or fabricated actions.
What to watch next
Watch whether visual platforms begin storing consent assertions with the generated asset and whether those records can later travel alongside C2PA provenance.
Sources
