The demonstrations are five seconds long and beautifully lit. The job is a forty minute interview recorded in a room with an air conditioner, a corporate cut that needs two shots nobody filmed, captions in three languages and a delivery deadline. This course puts AI tools where they earn their place in that job, and is equally clear about where they still waste your afternoon.
Day one is speech, which is where the reliable gains are. You transcribe difficult audio and assess the result honestly, including how these systems handle accents and Sri Lankan English, then cut a rough assembly straight from the transcript using text-based editing in Premiere Pro. You strip filler words and dead air without destroying the rhythm of someone's speech, and you generate, correct, style and burn in captions properly rather than accepting whatever the platform produces. The afternoon is audio repair and voice: recovering a badly recorded interview with speech enhancement and noise reduction, hitting loudness targets for broadcast, YouTube and social, generating a synthetic voice-over and working through the consent and disclosure rules that go with one. You test dubbing into Sinhala and Tamil and judge for yourself whether the current output is usable, since coverage of these languages lags well behind the major European ones.
Day two is generative video. You learn what a current model can and cannot hold steady across a shot, prompt for a specific frame by describing subject, camera move, lens, lighting and duration, extend a shot that is two seconds short using Generative Extend in Premiere Pro, and drive a shot from a still you art-directed yourself rather than from a text prompt alone. Cost and render time are treated as production facts: you budget a generated sequence with real numbers from the tools in front of you, because credits and render queues decide what is possible on a deadline.
The final afternoon is the hard part, which is making generated material sit next to filmed material without announcing itself. You match grain, colour, frame rate and motion, use AI-assisted rotoscoping, masking and object removal, and then cut a finished piece of sixty to ninety seconds that mixes filmed and generated shots. You finish with the Content Credentials attached to generated media, the disclosure rules the major platforms now apply to synthetic video and voice, and an export, archive and handover the client can act on.
Tools in this field change month to month, so the course teaches the judgement rather than a fixed button sequence: what to test, what to check before delivery and what to refuse. The class runs live online or in the classroom in Colombo, priced in LKR for Sri Lanka and in USD internationally.
