Complete-thought selection
A two-pass AI rubric scores hook, emotion, completeness, and quotability, and rule-based gates reject anything that ends mid-sentence or lands in the announcements.
Upload a full service recording. The pipeline transcribes every word, maps the service structure, and hands you one complete 20–90 second moment as a captioned vertical clip to review.
The pipeline transcribes every word, skips the welcome, announcements, worship, and offering, and proposes the strongest complete moments. You pick which ones become captioned vertical clips — nothing publishes without you.
The workflow starts with editorial clarity, so automation supports the sermon instead of flattening it.
A two-pass AI rubric scores hook, emotion, completeness, and quotability, and rule-based gates reject anything that ends mid-sentence or lands in the announcements.
Two bold lines per frame, cut from word-level timings, so captions never roll or repeat.
The selected moment arrives as a centered 1080×1920 H.264 MP4 with captions burned in, ready to review and download.
Choose an MP4, MOV, M4V, or WebM file up to 100 MB. Attach an SRT or VTT transcript if you have one; otherwise the pipeline transcribes the audio.
Only process media your church owns or has permission to reuse. Public access alone is not reuse permission.
Transcription, service mapping, moment selection, quality gates, pause snapping, captions, and the render each report progress. You can cancel at any point.
Watch the proposed moment with its source timestamps and excerpt, then download the MP4 or start another run. Nothing publishes automatically.
Practical answers for church communicators, editors, and ministry leaders.
It maps the service first, so welcome, announcements, worship, offering, and prayer segments are excluded. A two-pass AI rubric then scores the remaining moments for hook, emotion, standalone completeness, and quotability, and rule-based quality gates keep only 20–90 second windows that end on a finished sentence. When no AI endpoint is available, a rule-based engine selects instead and the result says which method ran.
The quality gates allow 20 to 90 seconds, and completeness matters more than a fixed target. Cut points snap to natural pauses so the clip does not start or end mid-word.
Transcription runs with a Bible vocabulary prompt, which helps with book names and theological terms, but it is not guaranteed. Verify every reference and caption against the source before you post.
Upload, word-level transcription, service-structure mapping, AI selection with rule-based gates, caption rebuilding, and the 1080×1920 render are live, with per-stage progress and cancel. Accounts, scheduling, brand templates, speaker tracking, and direct publishing are not built yet.
Upload one recording and review the moment the pipeline proposes.
Open the Clip Studio