Import (Setup)
On the Setup screen, drop MP4, MOV, or WAV files onto the dropzone — or use Select Files. Batch import is supported, so you can queue a whole scene's clips at once.
CineScribe Guide
DVP Tools / Post-Production
Automated Subtitles with CineScribe
Professional-grade transcription for filmmakers. Generate broadcast-accurate SRT and VTT files in seconds using OpenAI Whisper via fal.ai, complete with speaker diarization and QA tools.
Quick start
Subtitles in three moves
CineScribe works in two stages: a Setup screen where you import and configure, then an Editor where you refine and export. It drops you into the editor automatically once transcription finishes.
Configure & Generate
Still in Setup, choose your quality tier, language, and output (toggle Speaker Labels if you need them), then click Generate Subtitles. CineScribe uploads, transcribes, and lands you in the editor when it's done.
Refine & Export
In the editor, fix cues inline and check the QA tab for reading-speed and line-length flags. When it's clean, use the export bar to download SRT, VTT, TXT — or a PPTX supertitle deck for live performance. Use + Add media to head back to Setup for more clips.
Preparing your media
How to approach uploading clips
CineScribe uploads your file straight to fal.ai's CDN before transcription. There's no hard size cap in the app, but a little prep makes uploads faster and more reliable — especially for long projects.
Strip the video, send audio
Transcription only listens to the audio track. Export a WAV or MP3 and you'll upload a fraction of the size with identical accuracy — the single biggest speed win for any clip over a few minutes.
Split long runtimes
For features, long interviews, or anything past roughly an hour, break the media into reels or scenes and queue them as a batch. Shorter files upload faster, recover cleanly if one fails, and keep the cue editor responsive.
Supported formats
Video: MP4, MOV, MKV. Audio: WAV, MP3, M4A, AAC, FLAC, OGG. Drag several in at once — batch import queues them all.
Large files just take longer
Files over ~90 MB upload in chunks automatically, so a big clip won't fail outright — it simply takes longer on slower connections. If an upload stalls, use Retry Failed rather than re-queuing from scratch.
Rule of thumb
Audio-only + clips kept under an hour = the fastest, most reliable path. Reach for video upload only when you plan to burn subtitles back into the picture.
Methodology
The QA Pipeline
Subtitling isn't just about accuracy; it's about readability. CineScribe continuously monitors your cues against industry standard metrics.
The Four QA Stats
Every subtitle file is audited in real-time, highlighting exactly where viewers might struggle to keep up with the dialogue.
- Total Cues Absolute count of timecoded segments.
- Reading Speed Flags cues exceeding 20 characters per second.
- Timing Delta Identifies overlapping or zero-duration cues.
- Line Length Flags cues exceeding standard width limits.
fal.ai Integration
We use fal.ai's high-speed inference engine to run Whisper. This ensures that a 10-minute interview can be transcribed in under 30 seconds with 99% accuracy.
Tip
A fal.ai API key is required. Setup takes 2 minutes and includes free starter credits—perfect for testing the CineScribe pipeline.
Controls
Mastering the Workspace
Quality Tiers
The Quality selector trades speed for care — every tier runs Whisper large-v3 (there's no model-size choice). Draft and Swift use fal's fast wizper engine for the quickest turnaround. Balanced is the recommended default. Precision uses a smaller batch size for steadier chunking on difficult or noisy audio — slowest, but most careful.
Note on cue timing: the fast wizper tiers (Draft / Swift) segment coarsely — they're ideal for a quick, accurate full-text transcript but may group long passages into a single cue. For properly timed subtitle cues, use Balanced or Precision, then fine-tune in the cue editor.
Speaker Diarization
Toggle "Speaker Labels" to have the AI identify different voices. You can rename generic labels like "SPEAKER_01" to actual character names in the editor's Inspector tab.
Silence Filtering
Enable the "Silence Filter" to automatically suppress non-speech intervals, preventing the AI from hallucinating text during long pauses.
Inline Cue Editor
Click any text cell to edit instantly. The timecodes and speaker labels remain locked, allowing you to focus exclusively on linguistic accuracy.
Find & Replace
The fastest way to fix proper nouns. One pass can correct a character's name or technical term across 500+ cues simultaneously.
Session Persistence
Use Save Session (in the topbar Sessions menu) to store your full queue and edits in browser storage. Pick up right where you left off without re-uploading.
Professional Workflow
Post-Production Subtitling
Audio-Only Processing
Upload WAV or MP3 instead of large video files. It's 10x faster to upload and the transcription quality is identical to video processing.
The Raw Editor Bypass
Switch to 'Raw' mode to copy the entire subtitle block into an LLM (like Claude) for a creative rewrite or translation pass, then paste it back.
NLE Integration
Export SRT for Premiere or Resolve. If your editor supports it, use the 'Word Timings' export for even more precise subtitle placement.
QA Compliance
Never export a file with a 'Red' status chip. Resolving speed and length violations ensures your film remains accessible and professional.
Live Supertitles (PowerPoint)
Once you're in the editor, click Export Supertitles in the export bar for a one-click stage-ready deck — one cue per slide, large white text on black, with a blank slide between every line so the screen goes dark in the gaps. (You can also set the format to PPTX in the same bar and use Download Export.) Drop it straight into PowerPoint and let an operator advance each title live. For translated supertitles, transcribe with Output → English first so the slides carry the translation.