EU AI Act Requirements on Media: Images, Audio & Video

The EU AI Act places specific obligations on AI systems that generate, manipulate, or interact with media content — including images, audio, and video. These requirements primarily fall under transparency obligations and the regulation of deepfakes and all synthetic content.

Key Provisions Affecting Media Content

1. Deepfakes and Synthetic Media (Article 50(4))

The EU AI Act defines “deepfakes” as AI-generated or manipulated image, audio, or video content that “appreciably resembles existing persons, objects, places, or events” and would “falsely appear to be authentic or truthful”.

Requirements:

– Persons who deploy AI systems that generate or manipulate deepfake content “must disclose” that the content has been artificially generated or manipulated.

– This disclosure must be made “at the latest upon first publication or distribution”.

– The disclosure must be clear and distinguishable, informing the public that the content is AI-generated.

Exceptions:

– Content that is part of an “evidently artistic, creative, satirical, or fictional” work — though even here, disclosure must not undermine the rights and freedoms of third parties.

– Content used for “law enforcement” purposes where disclosure would compromise investigations.

2. AI-Generated Text on Public Interest Matters (Article 50(4))

When AI systems generate or manipulate text that is published for the purpose of “informing the public on matters of public interest”, it must be labelled as artificially generated or manipulated.

This has implications for AI-generated media captions, articles accompanying images/video, and synthetic news content.

3. Emotion Recognition Systems (Article 50(3))

AI systems that detect or infer emotions from “facial expressions, voice patterns, body language, or physiological signals” in images, audio, or video must:

– Inform individuals that they are being exposed to an emotion recognition system

– Provide this notification “prior to processing”

This applies to systems analysing video feeds, voice recordings, or images for emotional content.

Note: Emotion recognition in workplaces and educational institutions is prohibited under Article 5 (unacceptable risk), with very limited exceptions.

4. Biometric Categorisation Systems (Article 50(3))

AI systems that categorise individuals based on biometric data derived from images, audio, or video (e.g., inferring race, political opinions, religious beliefs, sexual orientation) must:

– Inform individuals that they are subject to such a system

– Comply with GDPR and other data protection requirements

Prohibition: Biometric categorisation systems that categorise individuals based on sensitive characteristics (race, religion, sexual orientation, political opinions) are banned under Article 5.

5. Real-Time Remote Biometric Identification (Article 5(1)(h))

The use of “real-time remote biometric identification” systems in publicly accessible spaces for law enforcement purposes is prohibited, with narrow exceptions:

– Targeted search for missing persons or abduction victims

– Prevention of a specific, substantial, and imminent terrorist threat

– Identification of suspects of serious criminal offences

This directly governs AI systems processing live video or image feeds for facial recognition.

6. Technical Labelling and Watermarking (Article 50(2))

Providers of AI systems that generate “synthetic audio, image, video, or text content” must ensure that:

– The outputs are “marked in a machine-readable format” that identifies them as artificially generated or manipulated.

– Technical solutions must be “effective, interoperable, robust, and reliable” as far as technically feasible.

– This includes the use of “watermarks“, metadata, cryptographic methods, or other machine-readable identifiers.

This obligation applies to the “provider” of the AI system (not just the deployer), meaning companies building generative AI tools must embed detection mechanisms into their outputs.

7. General-Purpose AI (GPAI) Models with Media Capabilities (Article 51-53)

Providers of “general-purpose AI models” (such as large language models or multimodal models) that can generate images, audio, or video must:

– Comply with transparency obligations including labelling synthetic outputs (meta-data, and other methods)

– Provide technical documentation about the model’s capabilities.

– Implement policies regarding copyright compliance for training data (including copyrighted images, music, and video).

– For “GPAI models with systemic risk”: conduct adversarial testing, assess and mitigate risks, report serious incidents, and ensure adequate cybersecurity.

Summary: AI Media-Related Activity & Required Obligations

Art. 50(4):  Image, Audio, Video
Deepfake generation/manipulation:
Requirement: Disclose AI generation to public.

Art. 50(2): Image, Audio, Video
Synthetic content generation
Requirement: Machine-readable labelling/watermarking

Art. 5(1)(h): Video, Image
Real-time biometric identification in public spaces
Requirement: Prohibited (with narrow exceptions)

Art. 50(3): Video, Audio, Image
Emotion recognition
Requirement: Inform individuals prior to processing

Art. 5(1)(g): Video, Image
Biometric categorization (sensitive attributes)
Requirement: Prohibited

Art. 50(3): Audio, Video, Image
Biometric categorization (non-prohibited)
Requirement: Inform individuals

Art. 50(4): Text (public interest)
AI-generated informational content
Requirement: Label as AI-generated

Art. 50(2), 51-53: All media types
GPAI model outputs
Requirement: Technical labelling + documentation

Who Is Affected?

These requirements apply to a broad range of organizations working with media:

– Generative AI providers (image generators, voice synthesizers, video creation tools)

– Social media platforms deploying AI content moderation or generation

– News organizations using AI for content creation

– Advertising and marketing agencies using AI-generated visuals or voiceovers

– Entertainment and gaming companies using synthetic media

– Surveillance technology providers using video/audio AI analysis

– Broadcasters using AI-generated or AI-enhanced content

Translation Implications for Media Compliance

Compliance with these media-related provisions creates specific translation needs:

– Transparency disclosures (“This content was generated by AI”) must be localized into the language of the audience in each EU member state.

– User notifications for emotion recognition or biometric systems must be provided in the local language.

– Technical documentation for generative AI systems must be available in required EU languages.

– Terms of use and policies governing AI-generated media must be translated for each market.

– Watermarking and labelling standards documentation may require multilingual versions for cross-border interoperability.

How Tongues Translation Services Can Help

We provide professional translation for all media-related EU AI Act compliance materials:

✅ Transparency and disclosure labels for synthetic content

✅ Terms of service and acceptable use policies for generative AI platforms

✅ Copyright compliance policies for training data

✅ Incident reporting documentation

✅ Public-facing communications about AI-generated content

✅ Support for all 24 official EU languages. Specialist AI and media terminology. 

EU AI Act: Deadline for Metadata and Watermarking Requirements

When Does the Watermarking and Metadata Obligation Take Effect?

The obligation for providers of AI systems to mark synthetic content with machine-readable metadata and watermarks takes effect on 2 August 2026.

This falls under Article 50(2) of the EU AI Act and applies as part of the broader transparency obligations for AI systems.

What Does Article 50(2) Require?

Providers of AI systems that generate synthetic audio, image, video, or text content must ensure that:

• Machine-readable marking: Outputs must be marked in a machine-readable format that signals they have been artificially generated or manipulated.

• Technical robustness:
  The marking mechanism must be effective, interoperable, robust, and reliable, as far as is technically feasible.

• Methods:
  May include watermarks, metadata, cryptographic methods, logging, fingerprinting, or other technical solutions as appropriate.

• Persistence:
  Markings should be resistant to removal, alteration, or circumvention, to the extent technically feasible.

• Accessibility:
  The marking must not compromise the usability, quality, or accessibility of the content.

• Scope of content:
  Applies to AI-generated or AI-manipulated images, audio, video, and text.

Who Does This Apply To?

The obligation falls on providers of AI systems — meaning the companies or entities that develop and place on the market generative AI tools, not just those who use them.

This includes providers of:

• Image generation tools (e.g., AI art generators, photo enhancement AI)
• Audio synthesis tools (e.g., text-to-speech, voice cloning, AI music generators)
• Video generation and manipulation tools (e.g., deepfake creators, AI video editors)
• Text generation tools (e.g., large language models, automated content generators)

Related Timelines and Context: Watermarking/MetaData

• 2 August 2025:
  GPAI model obligations take effect — providers of general-purpose AI models capable of generating synthetic content must begin complying with transparency and documentation requirements, including preparing for labelling obligations.
• 2 August 2026:
  Article 50(2) watermarking/metadata obligation fully applies — all providers of generative AI systems must ensure outputs are marked in machine-readable format.
• Ongoing (post-August 2026):
  European Commission may adopt implementing acts to specify technical standards for watermarking and metadata, potentially through harmonized standards or common specifications.

Stay Compliant. Stay Legal. Stay Ahead.

Tongues Translation Services enables EU firms to fully comply with the European AI Act while protecting their reputation and media.

Contact Tongues Translation Services Today

Let our compliance experts help you meet EU AI Act requirements with fast, accurate, and cost-effective solutions.