A new report from the British Film Institute (BFI) and the CoSTAR Foresight Lab, published in June 2025, outlines how generative AI is being adopted across the UK’s film, TV, and gaming.
Drawing on surveys, expert interviews, and public consultation, the report identifies a number of AI use cases across the screen sector, including several that relate directly to language-related applications — such as subtitling, dubbing, accent adaptation, and interactive dialogue generation.
At the BFI National Archive, large language models (LLMs), vision tools, and natural language processing methods are being used to make content more accessible and help users find relevant content more easily.
Current experiments include running speech-to-text on digitized videotapes using models like Whisper, WhisperX, and Nvidia’s Parakeet. The goal is to produce subtitle files (in WebVTT format) and pass transcripts into downstream AI services, including metadata enrichment and content tagging.
The team is also working with named entity recognition tools (spaCy), to identify relevant terms in subtitle and transcript text. These are then linked to Wikipedia or Wikidata — using tools like EntityFishing and ReFinED — supporting the creation of a semantically enriched search and discovery system for the national TV collection.
In parallel, the BFI is testing Google Gemini 1.5 Pro to generate automatic video descriptions, with plans to expand testing using open-source vision-language models such as Qwen2-VL and LLaVA-Video. The long-term goal is to connect all components — speech-to-text, entity linking, and video understanding — into a single pipeline running on in-house high-performance hardware.
Meanwhile, at the British Board of Film Classification, generative AI is being tested for deployment as part of its age ratings process. According to the report, models are being used to identify and tag potentially sensitive content — such as swearing, sexual references, or violence — to assist compliance officers with final classification.
AI Dubbing
In dubbing, London-based startup Flawless uses 3D facial tracking and generative models to synchronize actors’ lip movements with dubbed dialogue in multiple languages (“vubbing” or visual dubbing).
Generative AI is also being used to power interactive, unscripted dialogue in games and TV content. UK video games such as Dead Meat and 1001 Nights use LLMs to allow players to influence storylines and shape character dialogue dynamically.
Other use cases include voice refinement tools such as Respeecher, used to improve accent authenticity in scripted dialogue. Creatives are also turning to generative text editors like Grammarly to assist with grammar and tone consistency — particularly helpful for neurodivergent writers and those working in English as a second language. However, the BFI warns that overuse may lead to homogenized outputs and a loss of individual voice.
Slator 2025 Language Industry Market Report
The 150-page report offers a comprehensive view of the 2025 global market — with market sizing, AI capability breakdowns, buyer insights, use cases, survey data, and projections through 2030.
Impact on Writing and Translation
The report also highlights broader cultural concerns. Many generative models are trained on American English, raising questions about the dilution of British linguistic identity. Contributors call for more culturally representative training data, fine-tuning, and direct involvement from the creative community in model development.
While many in the UK screen sector view AI-driven efficiency gains as a positive development, the report raises concerns about the potential impact on the workforce. Writing and translation are among the skills increasingly being augmented — or replaced — by generative AI, according to the report, with entry-level roles seen as most affected.
At the same time, the shift is expected to create demand for new roles. The report highlights rising interest in professionals with skills in machine learning, language model fine-tuning, and AI tool development.