The shift is the result of a series of AI systems developed in-house and with external partners designed to standardise editing logic, automate clipping and structure visual storytelling, noted Lyn-Yi Chung, Deputy Chief Editor, Mediacorp.
Chung outlined a pragmatic approach to one of the most resource-intensive areas of newsroom work: video, at WAN-IFRA’s Jakarta AI Forum.
Chung, who spent a decade in television before moving into digital operations, described video production as “very painful to do” – dependent on specialised skills, multiple roles and tight turnaround times.
“Speed is everything. Timing is everything when it comes to video,” she said.
Mediacorp has built and deployed a set of AI tools designed to reduce production friction, formalise editing rules and accelerate digital publishing.
Chung highlighted three systems in particular: a story-structuring assistant known internally as “Visualiser GPT,” an automated editing tool called “AI Smart Frame,” and an AI-powered clipping system for broadcast bulletins.
Structuring follow-ups at speed
Visualiser GPT was designed to replicate the thinking process of an experienced video editor. Developed with input from a member of Mediacorp’s AI team, which has 35 years of production experience, the system is used during breaking news and for explanatory coverage.
“What sets the newsroom apart is the follow-ups that they can push out really quickly to help people make sense of an event,” Chung said.
Rather than simply generating text, the system suggests visual angles, identifies the types of footage required, proposes graphics and flags potential experts or secondary lines of inquiry.
It effectively lays out the structural skeleton of a visual story – what to show, who to speak to and how to frame the narrative for clarity.
The tool is also used to support newer newsroom staff in understanding how to construct visual explainers under deadline pressure.
Chung framed the system not as a replacement for editorial judgement, but as a way of systematising institutional knowledge – codifying years of production experience into a reusable system.
Teaching machines the grammar of editing
If Visualiser GPT addresses editorial logic, AI Smart Frame tackles execution.
Mediacorp trained the system using more than a year of previously edited television and digital footage.
The goal was to encode the conventions editors internalise over time: leading with a strong hook shot, sequencing long, medium and close shots naturally, and avoiding jarring cuts.
Editors upload raw clips alongside a script, and the system produces a rough cut that can be exported to standard editing software. Each suggested shot carries a confidence score.
“The confidence level has to be basically above 50 percent for us to consider it a decent choice,” Chung said.
Lyn-Yi Chung and Jason Subler, Global General Manager, Reuters.
Early iterations revealed that matching visuals line by line was insufficient.
“You need the NLP to figure out what the entire script means as a whole unit,” she noted. Development has therefore focused on holistic script comprehension rather than simple pairing.
The system remains in beta, with future phases aimed at automating captions and exporting multiple aspect ratios for cross-platform distribution.
In parallel, Mediacorp has deployed an AI system that automatically segments prime-time bulletins into digital clips.
Previously, editors could wait up to an hour for a segment to be prepared for online publication. Now, Chung said, clips can go live “within minutes,” with approximately 90 percent accuracy.
The system uses a combination of speech and visual cues to determine segment boundaries to determine boundaries – pairing introductions with corresponding packages or interviews.
The effect has been immediate.
When television journalists saw that their clips were being published faster and attracting more views, “people were excited and they readily embraced it,” she said.
Adoption, limits and culture
Chung acknowledged that not all aspects of video production lend themselves equally to automation.
Some processes, particularly audio refinement and dynamic reframing, still require human oversight.
More significant than technical constraints, however, is the challenge of adoption.
“People get testing fatigue,” she said. While journalists are eager for tools that save time, continuous scoring and evaluation can compete with daily production demands.
Her approach has evolved. Rather than requiring extensive weekly testing, she asks teams to map out all plausible use cases.
“As long as I know that I have an exhaustive list of scenarios, that’s all the commitment I need,” she said.
For Chung, the objective is neither novelty nor experimentation for its own sake. It is about reducing friction – compressing turnaround times, formalising tacit expertise and freeing editorial staff to focus on higher-value work.
Video, she argued, remains central to newsroom strategy.
At Mediacorp, AI is embedded in production workflows to make video faster, more predictable and operationally sustainable.