BFI report outlines rising use of generative tools across film, TV, and gaming
The UK’s film, television, and gaming industries are rapidly adopting generative AI to automate and enhance language-related tasks, according to a new report published this month by the British Film Institute (BFI) and the CoSTAR Foresight Lab.
Based on surveys, expert interviews, and public consultations, the report identifies a surge in applications ranging from subtitling and dubbing to dialogue generation, accent refinement, and accessibility enhancements.
Speech-to-text and AI-enhanced subtitling
At the BFI National Archive, teams are experimenting with automatic transcription of legacy media using speech-to-text models such as Whisper, WhisperX, and Nvidia Parakeet. These transcriptions are converted into subtitle files (WebVTT) and linked to downstream AI services for content tagging and semantic metadata generation.
Entity recognition tools like spaCy, EntityFishing, and ReFinED are also being used to link named entities in subtitles to Wikipedia and Wikidata entries — a move designed to build smarter search and discovery tools for the UK’s national television collection.
Dubbing with lip-sync accuracy
AI-driven dubbing — or “vubbing” — is another growing use case. London-based startup Flawless uses 3D facial tracking and generative models to synchronize actors’ mouth movements with dubbed dialogue in multiple languages. This enables more realistic multilingual adaptations without requiring full reshoots.
Tools like Respeecher are being employed to refine accents in scripted performances, allowing for more authentic voice work across regions and languages.
Games and interactive storytelling
Beyond traditional media, AI is being used to support unscripted, player-driven dialogue in interactive content. Video games such as Dead Meat and 1001 Nights incorporate large language models (LLMs) to allow players to shape conversations and storylines dynamically — a format gaining traction across UK studios.
The report notes rising interest in using generative AI to script branching narratives and create more immersive multilingual game environments.
Accessibility and classification
The BFI is also testing vision-language models such as Google Gemini 1.5 Pro, Qwen2-VL, and LLaVA-Video to automatically generate video descriptions for accessibility. The long-term goal is to connect transcription, tagging, and visual description tools into a single AI pipeline running on in-house hardware.
Separately, the British Board of Film Classification is exploring generative AI to assist in flagging sensitive content — including profanity, sexual references, and violence — to support age rating and compliance efforts.
Cultural and workforce implications
While the report acknowledges efficiency gains, it raises concerns over cultural and labor impacts. Many generative models are trained predominantly on American English, prompting fears that British linguistic identity could be diluted. Contributors urge greater investment in UK-specific data and direct creative oversight in model development.
The shift toward automation is also reshaping the workforce. Entry-level roles in writing and translation are among those most affected, while demand grows for professionals with expertise in AI, prompt design, and machine learning.
