Our active R&D project: AI presenters built from scratch for short-form video. Three problems, in order of difficulty.
One invented face that stays the same person across every frame and every video.
Natural head movement and clean lip-sync, checked frame by frame against the audio.
No real-person likenesses. No cloned voices. Every presenter is generated from scratch.
Render testing runs on cloud GPUs. Every attempt is logged for render time, cost and rejection reasons, so the pipeline gets cheaper and sharper with each pass.
The presenter work sits on top of the production system we already run every day: 51 owned short-form channels, every clip through automated technical checks and an editorial review before it can post, and reporting straight from platform analytics.
The Lab is the next layer of that system, and it is what our cloud-GPU render time is for.