ETSI Domain AI + Creative Media

Cross-domain marker for metaverse components that combine artificial intelligence capabilities with creative media applications such as generative content, procedural generation, and AI-assisted authoring.

Bridge-To

Semantic Classification

Content

Compositional Relationships (Components)

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:GenerativeContentClassification))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:AIArtCategorization))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:ProceduralContentGeneration))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:AIAssistedAuthoring))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:DigitalContentCreation))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:SyntheticMediaSystem))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:hasPart ai:AvatarCustomisationSystem))

Dependency Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:GenerativeAI))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:FoundationModel))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:RenderingPipeline))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:ContentPipeline))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:GPUCompute))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:ComputerVision))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:requires ai:DeepLearning))

Capability Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:GenerativeContentClassification))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:AIArtCategorization))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:VirtualWorldBuilding))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:SceneDesign))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:AvatarCustomisation))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:SyntheticMedia))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:ImageGeneration))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:VideoGeneration))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:enables ai:AudioSynthesis))

Implementation Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:implements ai:EUAIActTransparencyRequirements))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:implements ai:ETSIISGMECFramework))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:implements ai:C2PAProvenanceStandard))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:implements ai:GenerativeContentClassificationScheme))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:implements ai:AIArtTaxonomy))

Support Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:supports ai:GameDevelopment))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:supports ai:FilmProduction))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:supports ai:Metaverse))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:supports ai:AugmentedReality))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:supports ai:DigitalTwin))

Usage Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:uses ai:DiffusionModel))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:uses ai:GenerativeAdversarialNetwork))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:uses ai:LargeLanguageModel))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:uses ai:CLIPModel))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:uses ai:LatentSpace))

Reduction Relationships

SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:reducesTo ai:ETSIDomainAI))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:reducesTo ai:ETSIDomainCreativeMedia))
SubClassOf(ai:ETSIDomainAICreativeMedia
  ObjectSomeValuesFrom(ai:reducesTo ai:CrossDomainStandardisationMarker))

About

ETSI Domain AI + Creative Media occupies the intersection of two major work streams within the European telecommunications standards landscape: the broader ETSI Domain AI classification covering machine learning, neural network inference, and AI system deployment at the network edge, and the ETSI Domain: Creative Media classification covering 3D content authoring, rendering, multimedia production, and virtual world building. Together these parent domains describe respectively the intelligence substrate and the creative output domain; their intersection — captured in this cross-domain marker — identifies standards work items, technical requirements, and interoperability specifications that are specifically concerned with AI as a generative and transformative force within creative media production workflows for metaverse and spatial computing environments.

The domain emerged as a distinct standardisation category within ETSI’s metaverse taxonomy as Generative AI systems — particularly Diffusion Model and Large Language Models based architectures — began to displace or augment traditional Content Pipeline workflows in game development, virtual world authoring, broadcast media production, and immersive experience design. Where the ETSI Domain: Creative Media addresses the infrastructure and tooling requirements of creative media production (rendering pipelines, asset management, multimedia authoring), the AI + Creative Media cross-domain marker addresses the specific requirements that arise when AI systems take an active role in content generation: inference latency constraints at the ApplicationLayer, training data governance, Generative Content Classification for automated moderation of AI-generated user content, AI Art Categorization for provenance and attribution systems, Synthetic Media disclosure and watermarking, and the deployment of Foundation Model inference workloads within ETSI ISG MEC edge computing infrastructure.

The domain’s relationship to the EU AI Act is direct and consequential. Generative AI systems used in creative media contexts — text-to-image synthesis for advertising content, AI-generated synthetic voice for virtual assistants and characters, deepfake avatar animation — are covered by the EU AI Act’s Article 50 transparency obligations, which require providers to mark AI-generated outputs in machine-readable format. The C2PA (Coalition for Content Provenance and Authenticity) standard has emerged as the primary technical mechanism for EU AI Act-compliant content provenance, embedding cryptographically signed provenance manifests at the point of generation. ETSI’s standardisation work in this domain therefore encompasses not only the generative AI systems themselves but the provenance, attribution, and transparency infrastructure that surrounds them. The ETSI EN 304 223 baseline cybersecurity standard (2025) for AI systems includes provisions that apply to creative AI deployments, covering data security in training pipelines and the integrity of generative model outputs. An upcoming ETSI technical report (TR 104 159) will extend these principles specifically to generative AI, with particular attention to deepfakes, copyright and IPR concerns, and misinformation risks in creative media contexts — directly addressing the most contested aspects of the AI + Creative Media domain.

Historical Development and Context

The emergence of ETSI Domain AI + Creative Media as a distinct cross-domain classification reflects a broader trajectory in telecommunications standards history: the progressive incorporation of computational intelligence into what were previously content-agnostic infrastructure domains. In the earliest phases of ETSI’s metaverse taxonomy development, creative media and artificial intelligence were represented as entirely separate domain branches, with the Creative Media domain covering production tooling and rendering infrastructure, and the AI domain covering inference services and machine learning operations. The recognition that these domains were converging — that AI was not merely a service consumed by creative media production but was becoming the primary mechanism of creative media generation itself — drove the introduction of the cross-domain marker concept, of which AI + Creative Media is one instantiation alongside AI + Data Mgmt, AI + Governance, and AI + Human Interface.

The convergence accelerated dramatically from 2022 onward with the public availability of high-quality Diffusion Model systems. The release of Stable Diffusion in August 2022 as an open-weight model was particularly significant for the metaverse standardisation landscape: it demonstrated that foundation-model-scale generative capability could be deployed on consumer hardware without cloud infrastructure dependencies, enabling on-device and edge-deployed creative AI in metaverse contexts without mandatory cloud round-trips. This architectural possibility — edge inference for real-time creative content generation — is precisely the scenario that ETSI ISG MEC Phase 4 specifications must address, and it drove the explicit inclusion of creative AI content generation as a priority use case in ETSI’s metaverse edge computing work programme. ETSI GR ARF 012 V1.1.1 (2025) on Augmented Reality Framework extended the relevant scope to AR-specific creative AI applications: AI-driven scene understanding, AI-assisted content placement in AR overlays, and AI-generated augmentation of physical environments with virtual creative content.

The domain’s relationship to the broader ETSI metaverse taxonomy is navigational as well as technical: it serves as a cross-reference anchor enabling standards that touch both AI capabilities and creative media applications to be consistently indexed, avoiding the siloing of relevant standards within either the AI domain branch or the Creative Media domain branch alone. Standards authors working in ETSI technical committees can tag their outputs with the AI + Creative Media marker to ensure that specifications addressing AI-generated content for virtual world environments are discoverable both by creative media practitioners navigating the standards landscape and by AI system developers seeking standards applicable to their creative application deployments.

Components and Architecture

The ETSI Domain AI + Creative Media standardisation scope encompasses several interacting technical component families:

• Generative Content Classification: automated classification of AI-generated creative content by type, modality, provenance, and risk level; includes both human-initiated generation (text-to-image, text-to-video, text-to-music) and AI-augmented human-created content. Classification systems must interoperate with content moderation pipelines for virtual worlds and social platforms that host user-generated AI content. The EU AI Act’s two-tier taxonomy (fully AI-generated vs. AI-assisted) establishes the regulatory baseline; ETSI work must produce the technical specifications for implementing this classification at platform scale, including automated detection, metadata embedding, and audit trail generation.

• AI Art Categorization: systematic taxonomy and metadata frameworks for AI-generated artistic content, enabling attribution, licensing, copyright provenance tracking, and quality classification. Encompasses style classification (realism, abstraction, genre), generation method classification (diffusion, GAN, hybrid), and authorship attribution (fully automated, human-directed, co-generated). Relevant to digital art marketplaces, NFT platforms within virtual worlds, and advertising content libraries where AI-generated and human-created content must be distinguished for licensing purposes.

• Procedural Content Generation via AI: inference workloads that generate metaverse environment geometry, texture atlases, NPC dialogue, narrative branches, and ambient audio at runtime or near-runtime, scaling content availability beyond what human authorship can supply. Standards requirements include: inference latency budgets for interactive generation (sub-100ms for real-time integration with rendering pipelines), model serving architectures compatible with ETSI ISG MEC edge deployment, quality thresholds for generated assets relative to human-authored reference content, and content consistency protocols to ensure spatially coherent environments.

• AI-Assisted Authoring: tools and frameworks that integrate AI generation capabilities within traditional creative production workflows (3D modelling suites, game engines, video editing tools, audio DAWs). Standards requirements include: interface specifications for tool-level AI capability integration (plugin APIs, model serving protocols), provenance metadata propagation through creative pipelines, and quality evaluation frameworks for AI-assisted outputs relative to fully human-authored content.

• Synthetic Media disclosure and watermarking: technical mechanisms for identifying and labelling AI-generated content throughout the media lifecycle. Relevant standards include C2PA manifest schemas, SynthID-compatible watermarking specifications, and platform-level AI content labelling APIs. ETSI ISG MEC edge deployment specifications must include provisions for watermark embedding at the generation point, ensuring provenance chains are established before content enters distribution pipelines.

• Computer Vision for content analysis: AI-based scene understanding, object detection, and content classification systems deployed within creative media pipelines to enable automated quality assurance, content moderation, and metadata enrichment. In metaverse contexts, Computer Vision systems classify user-generated content, detect policy violations, and enable semantic search within large virtual environment content libraries.

  Use Cases and Major Families

  The cross-domain marker applies to several distinct deployment scenarios within metaverse and spatial computing ecosystems:

  Game content and virtual world generation: AI systems generate environment geometry, vegetation, architectural elements, texture variations, and ambient atmospheric content at scales that would require prohibitively large human authorial teams. Game engines such as Unreal Engine 5 and Unity 6 embed AI generation capabilities natively, with external model serving via APIs or on-device inference. Standards requirements include latency contracts for asset streaming into running game sessions, consistency and coherence specifications for procedurally generated environments, and quality thresholds relative to hand-authored reference assets. ETSI’s edge computing standards (ETSI ISG MEC Phase 4) directly enable low-latency AI inference for this use case by positioning inference workloads at the radio access network edge rather than in distant cloud data centres.

  Avatar and character generation and animation: AI systems generate and animate avatars, NPC characters, and synthetic performers within virtual environments. Capabilities range from static avatar appearance generation from reference images or text descriptions, to real-time facial animation from voice input, to autonomous NPC behavioural animation. Standards requirements include: avatar consistency specifications across metaverse platform boundaries, synthetic voice and facial animation interoperability protocols, and consent and identification frameworks for AI-generated human likeness.

  Creative media production for broadcast and advertising: Generative AI systems produce creative content assets (illustrations, concept art, advertising imagery, product visualisation, synthetic voice-over) within professional production pipelines. Standards requirements include: creative quality evaluation frameworks for AI-generated advertising content, Generative Content Classification for regulatory compliance (EU AI Act Article 50 transparency obligations), C2PA provenance manifest schemas for advertising media libraries, and licensing metadata propagation frameworks that preserve attribution through multi-step production workflows.

  AI-assisted musical scoring and audio production: Music generation systems (MusicGen, Suno, Stable Audio, ElevenLabs) produce ambient soundscapes, adaptive game music, and synthetic voice-over content for virtual environments. Standards requirements include: audio quality specifications for AI-generated music in immersive contexts (spatial audio compatibility, dynamic range requirements), copyright licensing metadata schemas for AI-generated music that distinguish between fully-licensed training data models and others, and disclosure requirements for AI-generated voice content under the EU AI Act.

  Content moderation and safety classification: AI systems classify user-generated content within metaverse environments to detect policy violations, illegal content, and harmful synthetic media (deepfakes, non-consensual intimate imagery). Automated classification must operate at scale and with latency compatible with real-time platform operations. Standards requirements include: classification taxonomy specifications, false positive and false negative rate thresholds, human review escalation protocols, and audit trail requirements under the EU Digital Services Act.

  Cross-Domain Relationships and Taxonomy Navigation

  Understanding ETSI Domain AI + Creative Media requires situating it precisely within the broader ETSI domain taxonomy and understanding the standardisation boundaries with adjacent domains. The cross-domain marker is intentionally not a fully independent domain but a tagged intersection point — a mechanism for ensuring that standards covering the AI-creative media convergence zone are discoverable from both parent domain branches.

  The principal adjacent domains and their boundary relationships with AI + Creative Media are:

  ETSI Domain: Creative Media (no AI qualifier): the parent Creative Media domain covering 3D content production, multimedia authoring, rendering pipelines, asset management, and scene design for virtual environments without an AI-specific focus. Standards exclusively addressing traditional (non-AI) creative media production workflows — rasterisation rendering specifications, asset format standards, content pipeline topology — fall in this domain but not in AI + Creative Media. Standards addressing AI-augmented versions of these same workflows — AI-assisted rendering optimisation, neural radiance field (NeRF) scene reconstruction, ML-based asset format conversion — fall at the intersection and carry both the Creative Media domain tag and the AI + Creative Media cross-domain marker.

  ETSI Domain AI (parent, without Creative Media qualifier): the parent AI domain covering all AI/ML capabilities in the metaverse taxonomy without creative media-specific scope. Standards addressing general AI inference infrastructure (ETSI ISG MEC AI service exposure, federated learning coordination, model serving specifications) fall in the AI domain but may not be specifically tagged AI + Creative Media unless they have specific creative media application requirements that differ from general AI inference. Standards addressing AI inference for creative content generation specifically — latency requirements for interactive image generation, quality thresholds for procedurally generated game content, Generative Content Classification schema specifications — fall at the intersection and carry both the AI domain tag and the AI + Creative Media cross-domain marker.

  ETSI Domain AI + Data Mgmt (sibling cross-domain marker): the parallel intersection of AI with data management. Creative AI systems have distinctive data management requirements — large creative training dataset management, generated content metadata management, provenance chain management for AI-generated creative artefacts — that overlap with the general AI + Data Mgmt scope. Standards addressing training data management for creative AI models (dataset curation for generative model training, diversity and bias assessment for creative training corpora) may carry both the AI + Creative Media and AI + Data Mgmt markers, reflecting the cross-cutting nature of data management requirements for creative AI systems.

  ETSI Domain AI + Governance (sibling cross-domain marker): the parallel intersection of AI with governance, ethics, and regulatory compliance. Creative AI governance — intellectual property compliance for training data, deepfake disclosure requirements, content moderation classification governance — overlaps with the general AI + Governance scope. Standards specifically addressing the governance of AI-generated creative content (attribution, copyright, disclosure) may carry both the AI + Creative Media and AI + Governance markers.

  ETSI Domain: Data Management + Creative Media: a further cross-domain marker in the ETSI taxonomy that covers data management requirements specific to creative media without the AI qualifier. Asset metadata management, content delivery network specifications, and digital rights management infrastructure fall here. When AI systems are specifically responsible for these data management functions in creative media contexts, the tripartite intersection of AI, data management, and creative media creates standards work items that may legitimately carry all three markers: AI, AI + Creative Media, and AI + Data Mgmt.

  The navigational value of the cross-domain marker taxonomy is thus that it enables standards discovery across multiple entry points: a creative media practitioner looking for standards applicable to AI content generation for their metaverse platform can navigate from Creative Media to AI + Creative Media; an AI system developer looking for standards applicable to their generative model deployment in a creative media context can navigate from AI to AI + Creative Media; and a regulatory compliance professional looking for all standards relevant to creative AI governance can navigate from AI + Governance to AI + Creative Media. This multi-entry-point discoverability is the primary functional purpose of the ETSI cross-domain marker architecture within the broader ETSI metaverse domain taxonomy.

  Evaluation Frameworks and Quality Benchmarks

  Assessing the quality of AI-generated creative content for standards compliance requires formal evaluation frameworks that go beyond informal aesthetic judgement. The absence of universally accepted cross-modal creative quality benchmarks is a significant open problem for the field, and ETSI standardisation in the AI + Creative Media domain must engage with both existing evaluation approaches and their known limitations.

  For image generation quality, the Fréchet Inception Distance (FID) and Kernel Inception Distance (KID) are the primary automated metrics, measuring the statistical distance between the distribution of generated images and real images in the feature space of an Inception neural network classifier. Lower FID indicates that the generated image distribution is statistically closer to the reference distribution. CLIP Score (or CLIPScore) measures the alignment between a generated image and its conditioning text prompt by computing the cosine similarity of their CLIP embeddings — a measure of semantic fidelity that complements FID’s distributional fidelity measure. Human Preference Score (HPS) and ImageReward train reward models on human aesthetic preference labels, providing a single score that correlates more closely with human quality judgements than FID or CLIPScore alone. For standards purposes, ETSI AI + Creative Media specifications must choose between these metrics or define composite evaluation profiles that combine automated metrics with human evaluation panels for holistic quality assessment.

  For audio and music generation, the Fréchet Audio Distance (FAD, analogous to FID for audio) and the Mean Opinion Score (MOS) from ITU-T P.800 provide the primary evaluation framework, with MOS in particular providing a well-established subjective quality measure from the telecommunications voice quality domain directly applicable to AI-generated speech and audio. For AI-generated music specifically, the MUSHRA (MUlti Stimulus test with Hidden Reference and Anchor) protocol provides a structured subjective evaluation framework used in audio quality research. These audio quality metrics have established precedent in ETSI’s telecommunications standards heritage, providing a natural bridge between the audio quality evaluation tradition and the AI-generated audio context.

  For 3D content generation (relevant to virtual world building and avatar generation use cases), evaluation metrics include Chamfer Distance and Earth Mover’s Distance for geometric fidelity, F-Score for geometric completeness, SSIM and PSNR for rendered appearance quality, and task-specific metrics (navigability, collision-free path existence, structural validity) for procedurally generated environments. The emergence of CLIP-based 3D quality metrics — evaluating 3D objects by rendering from multiple viewpoints and computing CLIP-based text-3D alignment scores — bridges the text-to-3D generation and text-to-image evaluation paradigms, though the accuracy and consistency of these metrics for 3D generation quality assessment remains an active research area.

  The standards implication of these evaluation frameworks is that ETSI AI + Creative Media specifications must include evaluation methodology annexes specifying which metrics apply to which content type, what reference datasets are used for FID/FAD computation, what sampling protocols govern human evaluation panels, and how composite scores are computed when multiple metrics are combined. The absence of agreed evaluation standards creates market fragmentation where different creative AI platforms report incomparable quality metrics, preventing informed procurement decisions and conformity assessment.

  Academic Context

  The academic research foundations of ETSI Domain AI + Creative Media draw from computational creativity, generative modelling, and standards engineering communities. Margaret Boden’s foundational framework of combinational, exploratory, and transformational creativity provides the conceptual vocabulary for evaluating what AI generative systems contribute beyond pattern recombination (Boden, 2004). Simon Colton and Graeme Wiggins developed the FACE, IDEA, and Creative Systems Framework models (2011–2019) that enable formal evaluation of computational creative acts — frameworks directly applicable to assessing whether AI systems operating within creative media domains satisfy creativity criteria or merely perform sophisticated interpolation.

  The convergence of Generative AI with metaverse technical requirements is explored in the ACM Transactions on Multimedia Computing paper “Unleashing Creativity in the Metaverse: Generative AI and Multimodal Content” (2025), which surveys the specific technical challenges of deploying generative AI systems within interactive virtual environments, including inference latency, content coherence, and multimodal consistency constraints. The Springer Nature review paper “Beyond Reality: The Pivotal Role of Generative AI in the Metaverse” (Kim et al., 2023) provides a comprehensive mapping of generative AI use cases within metaverse architectures. The PMC review “Generative Artificial Intelligence in the Metaverse Era: A Review on Models and Applications” (2025) surveys the latest model architectures and deployment patterns for AI-generated content in metaverse contexts, with sections specifically addressing standardisation gaps.

  Key research centres contributing to the academic foundations of this domain include the Creative Systems Lab (International Centre for the Study of Intelligence, University of the West of England / Monash), led by Simon Colton; Goldsmiths’ Autonomous Creative Systems group; Google DeepMind’s generative models team (responsible for Imagen, SynthID, and VideoPoet); Adobe Research’s creative AI group (Stable Diffusion, Firefly, ControlNet contributions); and the University of Edinburgh Informatics faculty’s generative model research. Industrial research increasingly dominates publication at NeurIPS Creative AI Track and CVPR, reflecting the compute intensity of frontier model development.

  The ETSI AI Conference 2025 (February, Sophia Antipolis) explicitly addressed generative AI standardisation gaps, including transparency requirements, conformity assessment approaches for creative AI systems, and the relationship between ETSI standards and EU AI Act implementation acts. The upcoming ETSI TR 104 159 on generative AI security will constitute the primary ETSI-authored technical reference for this domain’s standardisation scope.

  Key annual and recurring academic conferences relevant to this domain include:

• NeurIPS Creative AI Track: primary academic venue for creative AI research intersecting ML and artistic practice; formalised into conference proceedings from 2025; 2025 theme “Humanity” presented 93 papers and artworks

• International Conference on Computational Creativity (ICCC): annual conference for computational creativity theory, system design, and philosophy of machine creativity; active since 2010; primary venue for FACE, IDEA, and Creative Systems Framework models

• IEEE Conference on Computer Vision and Pattern Recognition (CVPR): primary venue for image generation quality, evaluation methodology, and technical architecture papers (Latent Diffusion Models, DreamBooth, ControlNet all published here)

• EvoMUSART: European conference on evolutionary and biologically inspired music, sound, art, and design; niche but historically important venue for algorithmic and evolutionary creative systems

• ISMIR (International Society for Music Information Retrieval): primary venue for AI music generation, evaluation, and analysis research (MusicLM, MusicGen, and music quality evaluation papers)

• ACM Multimedia: broad multimedia computing venue covering AI-generated video, multimodal generation, and creative applications in multimedia systems contexts

  Relevant academic journals:

• IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI): surveys of diffusion models, GAN variants, and image generation architectures

• ACM Transactions on Multimedia Computing, Communications, and Applications: applied multimedia AI research including metaverse content generation

• Artificial Intelligence Review (Springer): taxonomy and survey papers including the 2025 metaverse taxonomy review

• Journal of Artificial Intelligence Research (JAIR): computational creativity theory and evaluation framework papers

• Minds and Machines (Springer): philosophical foundations of machine creativity and the relationship between AI creativity and human cognition

  Benchmark Datasets and Evaluation Resources

  Standards work in the AI + Creative Media domain depends on access to benchmark datasets that enable objective comparison of generative AI systems. Key datasets and evaluation resources include:

  Image generation benchmarks:

• COCO (Common Objects in Context): 330K images with captions; used as reference distribution for FID calculation in text-to-image evaluation

• LAION-5B: open web-scraped dataset of 5.85 billion text-image pairs; the primary training data source for Stable Diffusion and many other open-weight generative models; its status under copyright law is contested in multiple jurisdictions

• DrawBench: a prompt benchmark of 200 text prompts designed to evaluate text-to-image generation on challenging spatial, counting, and compositional tasks

• T2I-CompBench: a compositional text-to-image generation benchmark evaluating attribute binding, spatial relationships, and non-spatial relationships

• GenAI-Bench: a multi-skill prompt benchmark evaluating generative AI systems on semantic coherence, photorealism, text rendering, and style adherence

  Audio and music generation benchmarks:

• AudioCaps: audio captioning dataset used for evaluating text-to-audio generation quality and semantic alignment

• MusicCaps: Google’s dataset of 5,521 human-annotated music clips used in MusicLM evaluation

• LibriSpeech: speech recognition benchmark used as reference distribution for evaluating AI-generated speech quality (MOS, WER)

• NSynth: a dataset of 300,000+ musical notes for evaluating timbre and instrument synthesis quality

  Video generation benchmarks:

• UCF-101: action recognition dataset used as reference distribution for video generation FID computation

• FETV: a fine-grained evaluation benchmark for text-to-video generation assessing temporal consistency, semantic adherence, and visual quality

• EvalCrafter: a benchmark for evaluating video generative models on multiple dimensions including visual quality, motion quality, temporal coherence, and text-video alignment

  3D content generation benchmarks:

• ShapeNet: a large-scale 3D shape dataset used as reference distribution for 3D generative model evaluation

• Objaverse: a large open dataset of 800K+ annotated 3D objects used for training and evaluating 3D generation models

• ABO (Amazon Berkeley Objects): a dataset of 147K products with 3D models and product metadata used for evaluating text-to-3D generation quality

  Current Landscape (2026)

  By mid-2026, the ETSI Domain AI + Creative Media standardisation landscape is shaped by several concurrent developments. The EU AI Act’s phased enforcement, with Article 50 transparency obligations for generative AI systems entering into effect, has driven demand for technical standardisation of content labelling and provenance mechanisms. The C2PA standard has been adopted by major creative AI platforms — including Adobe Firefly, Google, OpenAI, and Microsoft — establishing a de facto provenance standard that ETSI standardisation work now cross-references. The European Commission’s November 2025 Code of Practice on AI-generated content labelling established the policy framework within which technical standards operate, replacing the provisional two-tier taxonomy with a single uniform labelling regime. ETSI’s own EN 304 223 (2025) provides the baseline cybersecurity standard for AI systems applicable to creative AI deployments; TR 104 159 is in development to provide generative-AI-specific extension standards addressing deepfakes, copyright, and misinformation.

  The generative AI content market has bifurcated between licensed platforms (Adobe Firefly, ElevenLabs with Merlin/Kobalt licences, Getty Creative AI) commanding enterprise premium pricing, and unlicensed platforms subject to ongoing litigation. This market bifurcation is reflected in ETSI standardisation work through the emergence of training data provenance as a standards requirement alongside output provenance — a distinction between where the model learned from and what it produces. ETSI’s Phase 4 MEC work programme explicitly includes AI inference service deployment for metaverse applications, with creative AI content generation identified as a priority use case for edge inference deployment. The University of Manchester’s new AI research hub (opened 2024, £120 million investment) and the broader UK CoSTAR creative technology cluster infrastructure are building the research-to-standards pipeline that feeds into ETSI UK delegations.

  UK Context

  The UK occupies a distinctive position in the ETSI Domain AI + Creative Media standardisation landscape as home to world-leading creative industries worth £124 billion in GVA and significant AI research capacity. UK participation in ETSI standardisation activities in this domain is driven by the British Standards Institution (BSI), which serves as the UK’s national standards body and routes UK input into ETSI technical committees. The UK government’s Creative Industries Sector Plan (June 2025), part of the Modern Industrial Strategy, confirmed creative industries as a priority sector and committed £500 million in funding, including £100 million specifically for innovative companies developing creative technology — directly relevant to the AI + Creative Media domain.

  The Northern English creative cluster makes significant contributions to this domain through practice-based research and industrial deployment. The University of Manchester’s Alliance Manchester Business School and School of Computer Science study AI adoption in creative sector firms, with the Manchester digital economy and game development cluster providing industrial context. Newcastle University’s Digital Institute and the Northern Centre for Photography contribute to AI-assisted creative practice research. Leeds Arts University and Leeds Beckett University carry out fashion, design, and interactive media research with AI integration, while Sheffield DocFest and the Showroom Workstation provide festival and exhibition infrastructure that deploys creative AI tools in public cultural contexts.

  London and southern UK academic centres include: Goldsmiths, University of London (Autonomous Creative Systems group, Simon Colton and colleagues — foundational computational creativity research and practice-based AI art); UCL’s generative AI hub, which coordinates with Imperial College London, Cambridge, Oxford, Manchester, Edinburgh, and Surrey with partners including IBM, BT, DeepMind, and Cisco; the Royal College of Art (integrated AI creative tools across MA programmes in design, animation, and fine art); and Central Saint Martins (AI creativity from design and fashion practice perspectives). The Alan Turing Institute’s creative industries programme coordinates cross-institutional research with direct relevance to the standardisation domain. Scottish contributions come from the University of Edinburgh’s Informatics faculty (text generation, natural language generation, and computational creativity), the University of Glasgow’s historical strength in computational creativity research, and Rockstar North in Edinburgh — one of the world’s most commercially successful game studios and a major UK AI creative deployment site.

  The UK House of Lords Communications and Digital Committee’s 2024 report “AI, Copyright and the Creative Industries” (HL Paper 267) and the March 2026 UK Government Copyright and AI report (CP 2602959) together define the regulatory context within which UK contributions to ETSI AI + Creative Media standardisation are shaped. The UK’s position — maintaining copyright status quo rather than introducing a text and data mining exception — means that UK industry and academic contributions to ETSI standards in this domain carry a distinctive emphasis on provenance, attribution, and creator consent mechanisms that reflects the domestic policy environment.

  Major Variants and Standardisation Sub-Domains

  Within the overarching AI + Creative Media marker, several distinct sub-domain themes can be identified, each representing a cluster of related standards requirements:

  Generative image and visual content systems: the most commercially mature segment of creative AI in 2025–2026, encompassing text-to-image, image-to-image editing, super-resolution, inpainting and outpainting, and style transfer systems. Relevant standards requirements cluster around transparency and provenance (C2PA manifest schemas, EU AI Act Article 50 compliance), quality evaluation (FID, CLIP Score, human preference scoring), content safety filtering (preventing generation of illegal or harmful content from text prompts), and edge deployment (quantised model serving specifications for MEC edge nodes). The standardisation priority in this sub-domain is provenance and labelling infrastructure, reflecting the immediate regulatory requirements from the EU AI Act and the technical need for interoperable provenance metadata across major creative AI platforms.

  Generative audio and music systems: AI-generated music, voice synthesis, audio effects, and adaptive soundscape generation for metaverse environments. Standardisation priorities include audio quality evaluation (MOS, FAD, MUSHRA), copyright licensing metadata schemas for AI-generated audio (distinguishing licensed training data sources), consent mechanisms for voice cloning applications (regulatory requirement in multiple jurisdictions), and spatial audio compatibility specifications for AI-generated soundscapes in immersive metaverse contexts. The legal landscape for AI-generated music has evolved rapidly with Suno and Udio settlements with major labels in 2025, ElevenLabs Eleven Music licensing from Merlin and Kobalt, and the AI music market projected at 27.8% CAGR through 2034 — creating commercial urgency for standardised interoperability of AI music generation, licensing, and distribution infrastructure.

  Generative 3D and spatial content systems: AI systems generating 3D geometry, textures, materials, and spatial environments for virtual worlds and augmented reality overlays. Technically the most demanding and least mature segment, with specific standardisation requirements around geometric validity (collision-free geometry, navigable environments), consistency with existing virtual world environments (stylistic and structural coherence), and performance within game engine rendering pipelines (polygon budget, texture resolution, LOD generation). Standards work in this sub-domain connects to existing 3D content format standards (glTF, USD, OpenXR) and to ETSI’s ARF (Augmented Reality Framework) work, which addresses AI-generated spatial content in AR overlay contexts.

  AI-driven content moderation and classification: AI systems that classify, moderate, and govern AI-generated content within metaverse platforms. Standardisation priorities include classification taxonomy specifications (standardised content categories for policy violations, harmful synthetic media types), accuracy threshold specifications (false positive and false negative rate requirements for different content category severity levels), human review escalation protocols, audit trail specifications, and DSA (Digital Services Act) compliance architecture for very large online platforms. This sub-domain is operationally critical: without standards for AI content moderation classification, metaverse platforms operating across EU jurisdictions cannot demonstrate systematic compliance with DSA obligations for content governance.

  AI-generated avatar and digital human systems: AI systems generating, animating, and personalising virtual human representations for metaverse environments. Encompasses static avatar appearance generation (from text description, reference image, or body scan), real-time facial animation from voice or motion capture input, autonomous NPC behavioural animation, and digital human conversational agents. Standards requirements include avatar interoperability specifications (enabling avatar portability across metaverse platforms), synthetic likeness consent and identification frameworks, and AI-generated voice identification and disclosure specifications. The convergence of digital human generation with identity governance is a distinctive standardisation challenge — AI-generated avatars may represent real people (social VR contexts), fictional characters (game NPCs), or brand entities (virtual brand representatives), each requiring different identification and consent regimes.

  Future Directions (2026–2030)

  Several structural trajectories are expected to shape ETSI Domain AI + Creative Media standardisation over the 2026–2030 horizon:

  (1) Generative AI security standards (TR 104 159 and normative successors): The forthcoming ETSI TR 104 159 on generative AI security will establish the primary ETSI technical reference for generative AI-specific standards, covering deepfake detection, copyright compliance documentation, misinformation risk management, and confidentiality protections in creative AI pipelines. Normative standards derived from TR 104 159 will provide the harmonised standards baseline for EU AI Act conformity presumption in creative AI contexts, creating commercial incentives for platform adoption.

  (2) EU AI Act implementing acts pipeline: The EU AI Act’s implementing acts, expected through 2026–2028, will specify harmonised standards applicable to AI systems in creative domains — text-to-image generators, synthetic voice systems, video deepfake producers — creating a structured pipeline of ETSI standards work items. ETSI’s role as a designated ESO means these implementing acts will directly commission ETSI standardisation outputs, providing sustained funding and mandate for AI + Creative Media domain work.

  (3) C2PA and provenance infrastructure standardisation: The C2PA standard’s continued adoption by major platforms and its expected incorporation into ISO/IEC international standards frameworks will enable interoperable provenance across creative media supply chains spanning multiple jurisdictions and platforms. ETSI standards will cross-reference C2PA manifest schemas, establishing the technical vocabulary for EU AI Act-compliant content provenance in metaverse creative media contexts.

  (4) Edge inference deployment standards: ETSI ISG MEC Phase 4 and successor phases will evolve to accommodate the increasing scale and capability of generative AI workloads deployed at the network edge for real-time creative content generation. Standards will address:

  • Model quantisation specifications for edge-deployed generative models (latency vs. quality trade-off profiles)

  • Multi-tenancy specifications for shared edge AI inference supporting concurrent creative workloads

  • AI inference service exposure APIs enabling metaverse platform integration with edge-deployed generative services

  • Quality degradation profiles for graceful reduction of generation quality under compute constraint

    (5) Real-time and interactive generation standards: Latency Consistency Model (LCM) and flow matching architectures enabling sub-100ms image generation will require updated standards addressing interactive generation latency profiles, state consistency across generation frames, and real-time generation load balancing within edge computing infrastructure. Game engine integration specifications will address the APIs and protocols enabling generative AI systems to feed real-time content into running game and virtual world sessions without breaking frame rate targets.

    (6) Synthetic human likeness governance standards: Standards for AI-generated synthetic representations of real people — voices, faces, bodies — will address the identity governance requirements of digital human AI systems in metaverse environments. Relevant specifications will include:

  • Consent recording specifications for synthetic voice and likeness generation

  • Synthetic identity disclosure protocol specifications

  • Cross-platform avatar identity persistence specifications

  • AI-generated voice disclosure labelling schemas

    (7) Cross-platform AI content portability: As metaverse ecosystems proliferate and users move AI-generated content between platforms, standards for AI content metadata portability — preserving attribution, generation provenance, and licensing metadata across platform boundaries — will become essential. The Open Metaverse Interoperability Group (OMI) and related bodies provide adjacent standards context; ETSI AI + Creative Media work must align with or complement this cross-platform portability standards landscape.

    (8) Evaluation standard consolidation: The field’s current fragmentation between model-centric metrics (FID, CLIPSCORE), human preference metrics (HPS, ImageReward), and theory-driven creativity assessments (FACE, CSF) will consolidate around validated cross-modal evaluation frameworks as ETSI and ISO/IEC bodies publish normative evaluation methodology standards. Consolidated evaluation standards will enable objective conformity assessment for creative AI systems, supporting both regulatory compliance and informed procurement decisions.

    Formal Standards Landscape

    The standards ecosystem that ETSI Domain AI + Creative Media operates within spans multiple levels of abstraction, from high-level regulatory frameworks to specific technical protocol specifications. Understanding this landscape is essential for navigating the cross-domain marker’s scope and identifying which standards apply to a given creative AI deployment in a metaverse context.

    At the regulatory level, the EU AI Act (Regulation 2024/1689) is the primary binding instrument. For creative AI systems, Article 50 establishes transparency obligations for generative AI systems producing synthetic text, images, audio, and video: providers must ensure outputs are marked in machine-readable format and discernible as artificially generated. Systems that generate synthetic voices or manipulate existing images and video (deepfakes) face additional mandatory disclosure requirements. The Act’s general-purpose AI (GPAI) model provisions (Chapter V) apply to foundation models used in creative AI pipelines, requiring transparency documentation, copyright compliance summaries, and (for the most capable models) adversarial testing and incident reporting. ETSI’s role as a European Standards Organisation (ESO) means that ETSI standards can serve as harmonised standards under the EU AI Act, providing conformity presumption for in-scope AI systems — a commercially significant status that creates strong incentives for creative AI platform operators to align with ETSI standardisation outputs.

    At the ETSI technical standards level, the current authoritative instruments include: ETSI EN 304 223 V2.1.1 (2025), the baseline cybersecurity standard for AI models and systems, covering data security in training and processing pipelines, model integrity, and operational monitoring; ETSI TS 104 223 V1.1.1 (2025), the corresponding technical specification providing implementer guidance; ETSI TR 104 128 V1.1.1 (2025), the threat ontology for AI security applicable to creative AI deployments. The forthcoming ETSI TR 104 159 will be the first ETSI technical report specifically addressing generative AI, covering deepfakes, copyright and IPR compliance, misinformation risks, and confidentiality considerations — directly addressing the principal risk vectors of creative AI systems in metaverse and broadcast media contexts.

    At the cross-body coordination level, ETSI works in alignment with ISO/IEC JTC 1/SC 42 (AI fundamentals, vocabulary, bias, trustworthiness), CEN-CENELEC JTC 21 (EU AI Act harmonised standards pipeline), IEEE SA (AI ethics and autonomous systems standards), and ITU-T SG16 (multimedia and audiovisual systems, directly relevant to creative media AI). The C2PA open standard, developed by Adobe, Google, Microsoft, and other content industry partners, is increasingly referenced in ETSI work as the primary provenance mechanism for AI-generated creative content — a cross-body standards ecosystem reference that exemplifies how non-ETSI standards inform ETSI’s domain classification taxonomy.

    At the implementation specification level, the ETSI ISG MEC Phase 4 work programme addresses AI inference service deployment at the network edge, including inference latency requirements, API specifications for AI service exposure, and multi-tenancy specifications for shared edge AI infrastructure — all directly applicable to metaverse creative AI workloads that require edge-deployed generation to achieve real-time latency targets. ETSI’s ARF (Augmented Reality Framework) standards — specifically ETSI GR ARF 012 V1.1.1 (2025) — address AI-assisted scene understanding and content placement in AR contexts, contributing to the AI + Creative Media domain from the immersive technology angle.

    Technical Variants and Generation Families

    Creative AI systems within the ETSI AI + Creative Media domain span several distinct technical families, each with characteristic standards requirements:

    Text-conditional image generation (text-to-image): The dominant paradigm for AI creative content generation as of 2025–2026, implemented through Diffusion Model architectures (Stable Diffusion, DALL-E 3, Midjourney, Adobe Firefly, Imagen 3) that condition image denoising on text embeddings from CLIP or similar cross-modal encoders. Standards requirements specific to text-to-image systems include: cross-modal consistency specifications (alignment between text prompt meaning and generated image content), safety filter specifications (preventing generation of illegal or harmful content from text prompts), Generative Content Classification for automated output labelling, and C2PA manifest schema requirements for embedding generation provenance at the pixel level. Edge deployment requirements include model quantisation specifications for ETSI ISG MEC inference, specifying the accuracy-latency-power trade-offs acceptable for metaverse edge deployment.

    Latent video generation (text-to-video, image-to-video): Systems producing temporally coherent video sequences from text or image conditioning (Runway Gen-3, Sora, Kling, Pika Labs, Luma Dream Machine). Standards requirements include temporal consistency specifications (coherence of appearance, motion, and lighting across frames), audio-visual synchronisation specifications for systems that generate soundtrack alongside visuals, and disclosure specifications for synthetic video content under EU AI Act Article 50’s deepfake provisions. The computational intensity of video generation makes edge deployment challenging; standards for quality-of-service degradation profiles (reducing generation quality gracefully under compute constraint) are relevant for metaverse deployments where video generation must share edge compute with other workloads.

    Real-time interactive generation: Systems that generate or modify visual content at frame rates compatible with interactive experiences (60+ fps for VR, 120+ fps for high-performance gaming). Latent Consistency Models (LCMs) and flow matching architectures reduce diffusion sampling to 1–8 steps, enabling sub-100ms generation latency that makes interactive AI-generated content feasible. Standards requirements for interactive generation include: latency budget specifications distinguishing generation latency from rendering latency in mixed AI-rasterisation pipelines, state persistence specifications (ensuring generated content remains consistent across frames without complete regeneration), and computational resource reservation specifications within ETSI ISG MEC multi-tenant edge infrastructure.

    Procedural Content Generation (PCG) via machine learning: Systems that generate structured content (level geometry, narrative, NPC behaviour trees, quest templates) using ML models trained on large corpora of existing game or virtual world content. Unlike unstructured image generation, PCG systems must satisfy formal structural constraints (valid geometry, navigable level layout, coherent narrative structure). Standards requirements include: structural validity specifications for generated content, quality assessment framework specifications for ML-generated game content relative to hand-authored baselines, and versioning specifications for generative PCG systems used in live virtual environments (ensuring that content generated in the same environment at different times remains coherent and consistent).

    Neural Style Transfer and AI-assisted creative tools: Systems that apply learned aesthetic style to existing content or assist human creators in the generation process, operating within professional creative tool contexts (Adobe Photoshop Generative Fill, DALL-E inpainting, Stable Diffusion ControlNet). Standards requirements include: plugin API specifications for embedding AI generation capabilities in professional creative tools, provenance propagation specifications ensuring that AI-assistance events are recorded in C2PA manifests throughout complex multi-step creative workflows, and quality threshold specifications for AI-assisted outputs distinguishing AI-generated regions from human-authored regions in mixed-origin creative works.

    Governance, Ethics, and Regulatory Context for Creative AI

    The governance landscape for AI creative systems in ETSI’s metaverse taxonomy is shaped by a convergence of intellectual property law, content regulation, and platform governance requirements that together define the standards requirements for responsible creative AI deployment.

    Intellectual property and training data: The foundational governance challenge for creative AI systems is that training data for most commercially deployed systems was assembled from copyrighted human creative works. The EU Digital Single Market Directive’s text and data mining (TDM) exception (Article 4) permits TDM for commercial purposes unless rights holders have opted out — establishing an opt-out framework that applies to AI training data in EU member states. The UK has maintained the copyright status quo as of March 2026, providing no explicit exception and leaving training data legality governed by existing copyright law. These jurisdictional differences create compliance challenges for creative AI platforms operating globally; ETSI standardisation work must accommodate jurisdiction-specific requirements while establishing interoperable technical specifications. Training data provenance documentation — recording what data was used to train a model and under what licensing conditions — has emerged as a standards requirement that intersects with EU AI Act GPAI model documentation obligations and potential future regulatory requirements for training data disclosure. The ETSI AI + Creative Media domain therefore encompasses specifications for training data cataloguing, licence tracking, and provenance reporting as components of AI system technical documentation.

    Deepfake and synthetic human likeness: AI-generated synthetic representations of real people — voices, faces, bodies — present distinctive governance challenges that require specific standards provisions. The EU AI Act’s Article 50 requires specific disclosure for AI systems that generate or manipulate audio-visual content creating or appearing to be real persons; the UK Online Safety Act 2023 criminalises deepfake intimate imagery; US state-level legislation (California AB 2602, Illinois BIPA) establishes consent requirements for voice and likeness use. ETSI standards in this domain must specify the technical mechanisms for consent recording, synthetic identity disclosure, and detection of synthetic human likeness — areas where the C2PA standard’s AI identity and consent manifest schemas provide the primary technical vocabulary. The ETSI GR ARF 012 augmented reality framework addresses avatar identity in AR contexts, providing a related standards foundation for synthetic human likeness governance in spatial computing environments.

    Content moderation at scale: User-generated AI content within metaverse platforms — where users may generate, share, and remix AI content in real time — creates content moderation challenges that cannot be addressed through human review alone. AI-driven automated content moderation must classify and filter generated content at platform scale, detecting policy violations (violent content, illegal material, deceptive deepfakes) with latency compatible with real-time platform operations. ETSI Domain AI + Creative Media standards must specify the classification taxonomy, accuracy thresholds, and audit trail requirements for AI content moderation systems within metaverse platforms, operating within the framework established by the EU Digital Services Act (DSA) and the UK Online Safety Act. The intersection of generative AI (content creation) with AI moderation (content filtering) creates a recursive governance challenge: AI systems that generate content must be governed by AI systems that classify content, requiring clear standards for the accuracy, accountability, and transparency of both layers.

    Key Terminology

• Generative Content Classification: systematic categorisation of AI-generated media artefacts by type, provenance, modality, and risk level; the foundational classification function enabling regulatory compliance, content moderation, and attribution tracking in creative AI systems. Includes both automated ML-based classification and standardised metadata schemas for human-readable and machine-readable classification results.

• AI Art Categorization: a sub-domain of Generative Content Classification specifically concerned with classifying creative AI outputs within artistic taxonomies (style, genre, medium, period, cultural reference) and attributing them to specific generation systems, prompts, or training data sources. Supports digital art marketplace operations, NFT provenance, and licensing frameworks for AI-generated artistic works.

• Procedural Content Generation: algorithmic or ML-driven generation of structured content (levels, environments, assets, narrative) within defined formal constraints, enabling scalable content production for games, virtual worlds, and interactive experiences. Distinguishes itself from unstructured generative AI by the requirement for structural validity alongside aesthetic quality.

• Synthetic Media: media content (images, video, audio) produced or significantly altered by AI systems, including but not limited to deepfakes. Under EU AI Act Article 50, certain synthetic media must be disclosed as AI-generated. C2PA provides the technical standard for embedding this disclosure in machine-readable form.

• C2PA (Coalition for Content Provenance and Authenticity): an open technical standard for embedding cryptographically signed provenance manifests in media files at generation time, enabling verification of origin, edit history, and AI involvement. Increasingly adopted as the primary mechanism for EU AI Act-compliant content provenance in creative AI systems.

• Foundation Model: a large-scale model trained on broad multimodal data that serves as a base for specialisation via fine-tuning or prompting; the backbone of modern creative AI systems (Stable Diffusion, GPT-4V, Gemini, Claude 3 Opus), enabling diverse creative applications from a single pre-trained architecture.

• Latent Space: the continuous, compressed internal representation space learned by generative model encoders; navigation of latent space via prompting, interpolation, or optimisation enables controlled blending of styles, subjects, and aesthetic qualities in AI-generated creative content.

• CLIP (Contrastive Language-Image Pre-training): a cross-modal model trained on 400 million text-image pairs that aligns text and image representations in a shared embedding space; the foundational cross-modal architecture enabling text-conditioned image generation in diffusion models and AI art systems.

• SynthID: Google DeepMind’s AI-generated content watermarking system, embedding statistically robust, imperceptible watermarks in AI-generated images and audio to enable post-hoc detection; relevant to ETSI standards for AI content provenance and disclosure obligations under the EU AI Act.

• Diffusion Model: generative architecture implementing iterative stochastic denoising from Gaussian noise to high-quality sample; the dominant paradigm for image, video, and audio generation as of 2025–2026. Latent Diffusion Models (LDMs) operate in the Latent Space of a Variational Autoencoder, enabling high-resolution synthesis at tractable computational cost.

• Avatar Customization: the process of configuring the appearance, behaviour, and identity markers of a virtual avatar within a metaverse environment; AI-driven avatar customisation uses generative models to produce avatar appearances from text or image descriptions, automate avatar animation from voice or motion input, and personalise avatar aesthetics based on user preference models.

• AI-Assisted Authoring: creative production workflows in which AI generation capabilities are integrated within traditional authoring tools, enabling AI to assist, augment, or automate portions of the creative process while preserving human creative agency over the overall work. Distinguished from fully autonomous generation by the presence of continuous human direction and selection.

  Standards References and Key Documents

  The following are the primary ETSI and cross-body standards documents within scope of ETSI Domain AI + Creative Media:

  ETSI primary documents:

• ETSI EN 304 223 V2.1.1 (2025-12): Securing Artificial Intelligence (SAI) — Baseline cybersecurity requirements for AI models and systems

• ETSI TS 104 223 V1.1.1 (2025-04): Securing Artificial Intelligence (SAI) — Technical specification

• ETSI TR 104 128 V1.1.1 (2025-05): Securing Artificial Intelligence (SAI) — Threat ontology

• ETSI TR 104 159 (forthcoming): Securing Artificial Intelligence (SAI) — Generative AI specific requirements

• ETSI GR ARF 012 V1.1.1 (2025-11): Augmented Reality Framework (ARF) — AI content in AR contexts

• ETSI GR MEC-DEC 050 V4.1.1: MEC Deployment — Metaverse and edge AI service deployment

• ISG MEC Terms of Reference v2.5 (October 2024): Phase 4 scope definition for AI-native MEC

  Cross-body standards in scope:

• C2PA Technical Specification 2.0 (2024): Content provenance and authenticity for AI-generated media

• ISO/IEC 22989:2022: Artificial intelligence — Concepts and terminology

• ISO/IEC 42001:2023: Artificial intelligence management system

• ISO/IEC TR 24027:2021: Bias in AI systems and AI-aided decision making

• EU Regulation 2024/1689 (EU AI Act): Articles 50 (transparency), Chapter V (GPAI models)

• EU Directive 2019/790 (DSM Directive): Articles 3–4 (text and data mining exceptions)

• UK Online Safety Act 2023: Deepfake intimate imagery offences

• UK Data (Use and Access) Act 2025: Sections 135–137 AI training data review

  Key ETSI technical reports and white papers:

• ETSI White Paper No. 64: AI Technologies in ENI to Increase Autonomous Operation

• ETSI AI Conference 2025 Proceedings: Generative AI standardisation sessions

• ETSI GR SAI series: Securing AI application domain reports (by sector)

  Research and Literature

  1. Kim, J., Park, J., & Cho, H. (2023). Beyond reality: The pivotal role of generative AI in the metaverse. arXiv:2308.06272.
  2. Li, Y., Zhang, J., & Wang, M. (2025). Unleashing creativity in the metaverse: Generative AI and multimodal content. ACM Transactions on Multimedia Computing, Communications, and Applications. https://doi.org/10.1145/3713075
  3. ETSI. (2025). ETSI EN 304 223 V2.1.1: Securing Artificial Intelligence (SAI) — Baseline cybersecurity requirements for AI. European Telecommunications Standards Institute.
  4. ETSI. (2025). ETSI TS 104 223 V1.1.1: Securing Artificial Intelligence (SAI). European Telecommunications Standards Institute.
  5. ETSI. (2025). ETSI TR 104 128 V1.1.1: Securing Artificial Intelligence (SAI) — Threat ontology. European Telecommunications Standards Institute.
  6. ETSI. (2024). ISG MEC Terms of Reference v2.5. ETSI Portal Document. https://portal.etsi.org/Portals/0/TBpages/MEC/Docs/ISG_MEC_ToR_Approved_20241014.pdf
  7. ETSI. (2025). ETSI GR ARF 012 V1.1.1: Augmented Reality Framework (ARF). European Telecommunications Standards Institute.
  8. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. CVPR 2022, 10684–10695.
  9. Radford, A., Kim, J.W., Hallacy, C., et al. (2021). Learning transferable visual models from natural language supervision (CLIP). ICML 2021.
  10. Ho, J., Jain, A., & Abbeel, P. (2020). Denoising diffusion probabilistic models. NeurIPS, 33.
  11. Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. (2014). Generative adversarial nets. NeurIPS, 27.
  12. Boden, M.A. (2004). The Creative Mind: Myths and Mechanisms (2nd ed.). Routledge.
  13. Colton, S., Charnley, J., & Pease, A. (2011). Computational creativity theory: The FACE and IDEA descriptive models. ICCC 2011.
  14. Wiggins, G.A. (2019). The problem of creative computing. In A Guide to the Creative Potential of Computers, 7–35.
  15. European Commission. (2025). Commission launches work on a code of practice on marking and labelling AI-generated content. Digital Strategy Factsheet.
  16. European Parliament Research Service. (2025). Generative AI and copyright: Training, creation, regulation. PE 774095.
  17. UK House of Lords Communications and Digital Committee. (2024). AI, Copyright and the Creative Industries. HL Paper 267.
  18. HM Government. (2026). Report on Copyright and Artificial Intelligence. CP 2602959.
  19. HM Government. (2025). Creative Industries Sector Plan. Modern Industrial Strategy.
  20. Dhariwal, P., & Nichol, A. (2021). Diffusion models beat GANs on image synthesis. NeurIPS, 34.
  21. Ruiz, N., Li, Y., Jampani, V., et al. (2023). DreamBooth: Fine-tuning text-to-image diffusion models for subject-driven generation. CVPR 2023.
  22. Croitoru, F.A., Hondru, V., Ionescu, R.T., & Shah, M. (2023). Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence.
  23. Agostinelli, A., Denk, T.I., Borsos, Z., et al. (2023). MusicLM: Generating music from text. arXiv:2301.11325.
  24. PMC. (2025). Generative artificial intelligence in the metaverse era: A review on models and applications. PMC Article 12364526.
  25. Springer Nature. (2025). Groundbreaking taxonomy of metaverse characteristics. Artificial Intelligence Review. https://doi.org/10.1007/s10462-025-11243-5
  26. ETSI AI Conference. (2025). Session proceedings: Generative AI, AI methods and related standardisation. Sophia Antipolis, 10–12 February 2025.
  27. AI Standards Hub. (2025). ETSI/ISG ENI archives. https://aistandardshub.org/search_committee_reference/etsi-isg-eni/
  28. Coalition for Content Provenance and Authenticity (C2PA). (2024). C2PA Technical Specification 2.0. Content Authenticity Initiative.

  Interoperability Considerations

  Cross-platform interoperability for AI-generated creative content is one of the most operationally significant challenges in the ETSI AI + Creative Media domain. Unlike human-authored creative content, AI-generated content carries additional metadata — generation provenance, model version, training data category, prompt history, safety filter audit records — that must be preserved and transmitted across platform boundaries to support regulatory compliance, attribution, and quality assurance workflows.

  Key interoperability dimensions include:

• Content format interoperability: AI-generated images, video, audio, and 3D content must be encoded in standard formats (JPEG XL, WebP, H.265, FLAC, glTF 2.0, USD) that strip or separate AI provenance metadata unless explicit embedding mechanisms (C2PA XMP manifests, IPTC AI metadata schema) are used. Standards must specify how AI provenance metadata survives transcoding, editing, and format conversion operations.

• Provenance metadata interoperability: C2PA manifests provide the primary technical standard for embedded provenance, but different AI platforms produce manifests with varying levels of detail. ETSI standards must specify minimum manifest content requirements for cross-platform provenance interoperability in metaverse content supply chains.

• Content classification interoperability: AI content moderation classifiers used by different metaverse platforms may produce incompatible content category labels. Standards for content classification taxonomy interoperability — enabling classification results from one platform’s AI to be interpreted by another platform’s content policy enforcement system — are essential for federated metaverse content governance.

• Avatar and digital human portability: AI-generated avatars must be exportable in interoperable formats (glTF 2.0, USD, VRM) with preservation of AI generation metadata, consent records, and identity governance information. The ETSI ARF framework and the emerging Avatar Interoperability Profile (AIP) initiative provide the relevant standards context.

• Model API interoperability: Creative AI services exposed via APIs within ETSI ISG MEC edge infrastructure must support interoperable API schemas enabling metaverse platform software to switch between AI content generation services from different vendors without application-level changes. OpenAPI-based AI service interface specifications are the primary vehicle for this interoperability requirement.

• Safety filter interoperability: AI content safety filters applied at generation time must produce machine-readable safety assessment metadata in interoperable formats, enabling downstream content governance systems to interpret and act on safety assessments from multiple creative AI generation services without bespoke integrations for each provider.

• Licensing metadata interoperability: AI-generated content deployed in commercial contexts requires interoperable licensing metadata schemas enabling downstream users to verify what training data was used, what licensing terms apply to the generated content, and what attribution obligations are imposed by the applicable licences. IPTC’s AI-generated image metadata standards and the Structured Data on the Web schema for AI content licensing provide the primary reference points.

• Quality metric reporting interoperability: Creative AI platforms must report content quality metrics in interoperable formats enabling multi-platform quality comparisons and objective procurement decisions. Standardised metric reporting schemas (specifying how FID scores, CLIP scores, and MOS ratings are computed and reported) will emerge as ETSI and ISO standards, enabling independent quality verification for creative AI service procurement.

• Generation history portability: In co-creative workflows where human creators and AI systems iteratively develop creative content, the history of generation steps (prompts, parameters, model versions, intermediate outputs) must be portable across tools and sessions in interoperable format. This generation history portability enables iterative refinement workflows to resume in different tools, facilitates credit attribution, and supports content provenance documentation under the C2PA standard.

  See Also

  Related ETSI domain entries that should be consulted alongside this entry:

• ETSI Domain: Creative Media — the parent domain covering creative media without the AI qualifier

• ETSI Domain AI — the parent AI domain covering all AI capabilities in the metaverse taxonomy

• ETSI Domain AI + Data Mgmt — the sibling cross-domain marker for AI and data management

• ETSI Domain AI + Governance — the sibling cross-domain marker for AI and governance

• ETSI Domain AI + Human Interface — the sibling cross-domain marker for AI and human interface

• ETSI Domain: Data Management + Creative Media — data management specific to creative media contexts

• ETSI Domain Taxonomy — the complete ETSI domain classification framework

• Creative AI — the broader ontology entry for AI-driven creative systems

• Generative AI — the parent technology class enabling creative AI capabilities

• Computer Vision — cross-reference AI bridge for image-based creative applications

Provenance