Caption workflows
From transcript data to captioned video.
Run speech recognition in your own workflow, map the approved text into a video composition, and collect the finished MP4 asynchronously.
Keep transcription and rendering separate
Whisper or Deepgram can produce transcript text and timing in an upstream service. Your workflow can normalize that output, prepare supported composition variables, and submit a Rendofy render request. Rendofy is the render step, not a speech-to-text endpoint.
Send audio to your chosen speech-recognition provider and retain its transcript response.
Map text, media, and styling inputs to a template that has been tested for your output.
Submit JSON and use the callback's MP4 URL in the next workflow step.
Example of an upstream transcript object
This is an illustration of data your workflow might normalize. It is not a documented Rendofy caption schema. The public intake contract uses api_key, callback_url, and a structured payload.
{
"transcript": {
"text": "A new collection, ready for the weekend.",
"segments": [
{ "start": 0.0, "end": 1.2, "text": "A new collection" },
{ "start": 1.2, "end": 2.4, "text": "ready for the weekend." }
]
}
}The current public API documentation does not specify direct SRT ingestion, per-word timestamp styling, font or color override fields, or a guaranteed 9:16 output preset. Confirm those details with the selected composition before production use.
Treat caption styling as composition behavior
For word-level emphasis, map each transcript segment's start time, end time, and text into a composition that explicitly supports timed words. Set font, color, and safe-area rules in that composition, then confirm the exported aspect ratio. These are implementation requirements for a tested composition, not universal Rendofy request fields.
Validate layout before batching
For vertical social output, verify that your composition and source assets produce a 9:16 frame, that caption safe areas avoid platform controls, and that audio timing remains aligned. Test a single callback and inspect the actual MP4 before increasing batch volume.