AI-Powered Production
Multimodal AI woven through the production chain.
Mingmou multimodal understanding and DeepSeek across planning, production, and live signals — from APE 2.0 to the AI200 analysis unit.
Overview
The AI-Powered Production solution threads one intelligence layer through the whole chain: APE 2.0 plans, writes, and assembles with Mingmou multimodal understanding and DeepSeek; Media Hive enriches and retrieves across the library; and the Mediavate AI200 tags live signals in real time.
Challenge
AI lands in broadcasting as disconnected point tools — a transcription service here, an image generator there — each with its own storage, formats, and integration project. Intelligence that doesn't flow into the media's metadata dies where it was computed.
Point-tool sprawl
Separate AI services for speech, faces, and generation multiply vendors without compounding value.
Live-signal blind spot
Most AI runs on files after the fact; live production needs recognition while the signal is still on air.
Tags that go nowhere
Analysis results stranded in dashboards never reach the editors who could cut with them.
Generation guardrails
AIGC belongs inside the production workflow — with editorial review — not in a browser tab beside it.
Solution
Sobey applies one AI capability set at three altitudes.
- In planning and production — APE 2.0 runs story-lead agents, AI script writing with review gates, full-library media analysis, SOT matching, and full-modal AIGC — digital humans, image, video, and music generation with ComfyUI workflow support.
- In the library — Media Hive enriches every asset with speech-to-text, face recognition, and shot and label detection, and offers Bring-Your-Own-AI for organizations with preferred engines.
- In the signal chain — The AI200 analyzes up to 20 live channels with under 2 seconds average latency — faces, scenes, speech, entities, shot motion — writing tags directly into material metadata files that post-production reads for intelligent matching, grading, and voice editing.
Frame-level tagging makes all of it retrievable.
Result
Intelligence compounds instead of fragmenting: what the AI sees on the live signal, the library remembers, and the newsroom cuts with — one capability set from ingest to air.
Live channels analyzed in real time by a single AI200 unit.
Average latency from live signal to written metadata tags.
AIGC modes — digital human, image, video, music, and broadcast graphics.
Explore every solution.
News, sports, studio, remote, playout, archive, AI — see how Sobey builds each workflow end to end.