AWS MediaConvert adds AI subtitles to video transcoding jobs
AWS added Smart Subtitles to MediaConvert on November 6, 2026, bringing AI transcription into the same job that transcodes and packages an on demand video. Publishers can generate captions while preparing distribution files, rather than maintaining a separate caption generation pipeline.
The useful change is where the work happens. It gives teams already using MediaConvert another way to assemble a video delivery workflow. Whether it saves time overall will depend on the audio, the output requirements and how much correction the captions need.
Six languages, with a defined regional rollout
The launch supports English, French, German, Italian, Portuguese and Spanish. AWS lists five regions: N. Virginia, Oregon, Ireland, Mumbai and Sydney. Outputs can use separate text caption files or captions burned into the video, subject to format compatibility.
AWS describes the underlying feature as same language subtitle generation. Support for several source languages should therefore not be read as a promise to translate a video into all of them. A publisher planning translated versions still needs to establish a separate translation workflow.
This is also an extension of existing technology. AWS introduced Smart Subtitles for live MediaLive broadcasts in May 2026. The November announcement puts the capability inside MediaConvert’s on demand processing workflow.
What a MediaConvert job needs
The MediaConvert setup guide starts at the output, not a new standalone transcription project. Add an output group and output, add captions in the encoding settings, then choose Smart subtitle as the captions source. Select a compatible destination type, the audio source and a language that matches that audio.
There is an important input constraint in the MediaConvert API reference: the audio selector chosen for Smart Subtitles must exist in every input in the job. AWS also says it cannot generate subtitles when it cannot detect speech.
For a job combining several clips, check each input’s audio selection before starting a batch. A correctly named selector in the first clip does not establish that the remaining clips supply the intended dialogue. Keep a short representative job as a configuration check before processing an archive.
Dictionaries help with names, but still need review
AWS supports custom dictionaries for proper nouns, brands and specialized terminology. The Elemental Inference feature settings identify the dictionary as an optional input intended to improve domain specific transcription. They also distinguish a required source language from optional layout and filtering settings.
Those settings include English regional variants and optional profanity handling. The documented filtering choices preserve the transcript, replace profanity with asterisks or remove it. The guide limits that filter to English, Spanish and Italian. Check which controls your MediaConvert workflow exposes rather than assuming every Elemental Inference option appears in the same place.
For an interview series, a small dictionary of recurring speakers and product names is a sensible starting point. Review the resulting captions against the actual audio. A dictionary can guide recognition, but it does not establish that a particular sentence or name was transcribed correctly.
Evaluate the finished captions, not just the job status
A useful trial should include quiet dialogue, background music, overlapping speech and the accents in the intended catalog. Compare captions with the spoken words, then inspect their timing, line breaks and placement in the player or burned video. These are suggested acceptance checks, not tests ByteForward has performed.
Our coverage of Hugging Face’s expanded English accent evaluations explains why aggregate speech recognition scores need context. That leaderboard is not an evaluation of this AWS feature, but the same selection principle applies: use recordings representative of the people and conditions you need to support.
Budgeting needs a separate check. The official MediaConvert pricing and Elemental Inference pricing pages reviewed for this article did not establish a specific on demand Smart Subtitles rate. Live channel examples should not be copied into an estimate for a MediaConvert job.
For existing MediaConvert users, the feature is worth a controlled trial where caption generation is currently a separate step. Measure processing cost and editorial correction time together. The integration is most useful when it reduces the work needed to deliver captions that are ready for viewers.
