This guide explains how to use the portal interface after login. The full operations manual remains available only to admins.
Library page
Upload source videos with drag-and-drop or the file picker.
Review detected file details such as resolution, codecs, duration, bitrate, and storage path.
Use Convert to create a new MP4, HLS, or combined job.
Use Download source to download the original uploaded file.
Jobs page
Monitor queued, running, completed, canceled, or failed jobs.
See requested output settings, encoder path, progress, and generated asset sizes.
Preview MP4 or HLS in the browser when outputs are ready.
Download MP4 or HLS ZIP files from the completed job actions.
Choosing output mode
MP4 only shows only MP4-related fields.
HLS only shows only HLS ladder fields.
MP4 + HLS shows fields needed for both outputs.
When multiple HLS variants are created, the preview uses the master playlist for adaptive playback.
Storage and limits
Your storage pill in the top bar shows used, total, and utilization percentage.
Quota counts original uploads and generated outputs together.
If an upload or conversion would exceed your remaining space, the action is blocked.
Delete old source files or completed jobs to recover space when needed.
Transcripts page
Click Transcribe next to any uploaded file in the Library.
Pick a Whisper model, source language (or auto-detect), and whether to translate to English.
Choose SRT/VTT downloads, a burned-in MP4, or both.
The server uses GPU acceleration automatically when available (NVIDIA CUDA on Linux, Apple Metal on macOS) and falls back to CPU on failure.
Open the Transcripts tab to see queued, running, and completed jobs with progress bars.
Preview behavior
MP4 previews play directly in the browser. HLS previews use native playback on Safari and HLS.js on other supported browsers. When adaptive bitrate variants exist, the quality details panel lists the ladder so you can confirm ABR outputs were generated.
Queue control
This version removes the former hard cap of 5. Use a value that matches your CPU, RAM, disk, and GPU capacity.
How many transcription jobs run in parallel. Whisper is heavy — start with 1 and increase only after observing GPU/RAM headroom.
0 = auto from CPU count and concurrency. Whisper's per-chunk thread scaling plateaus near 8.
0 = auto. Splits each job's audio into N chunks processed in parallel. The throughput winner on many-core CPU boxes.
System metrics
Loading…
Transcription engine
Loading…
Users
User
Role / channels
Quota
Actions
Create conversion job
When NVIDIA NVENC is available the server will try GPU encoding first and transparently fall back to CPU encoding if needed.
One per line: widthxheight:bitrateKbps:name. When more than one variant exists, the HLS master playlist exposes multiple ABR bitrates for native Safari playback and HLS.js playback.
Transcribe video
Larger models are more accurate but slower and use more RAM/VRAM.
"Auto-detect" lets whisper pick the language from the first 30 seconds of audio.
Whisper can translate any supported language to English in a single pass.
Burning re-encodes the source. Plain SRT/VTT downloads do not.
Both formats are produced together in a single whisper pass.
Only used when burn-in is enabled.
0 = use the admin default. Whisper's per-chunk thread scaling plateaus near 8.
0 = use the admin default. Splits the audio into N chunks and runs them concurrently. Best speedup on many-core CPU boxes.
t × p ≈ cores used per job
Preview
Adaptive bitrate details
For Safari on macOS, native HLS playback chooses the rendition automatically.