Configuration Reference

Overview

Global settings are stored in config.yaml in your Onset Engine data folder, %APPDATA%\Onset Engine\config.yaml (paste that path into the File Explorer address bar to open it). It is not in the installation folder. Close Onset Engine before editing it. These values serve as defaults - most can be overridden per project in the Editor.

Full Reference

editor:
  audio_crossfade: 2.0          # Crossfade between consecutive TRACKS in a playlist (seconds)
  ducking_volume: 0.3           # Music level while your clips' audio is mixed in (0.0–1.0)
  fade_in_dur: 1.0              # Fade-in duration at video start (seconds)
  fade_out_dur: 2.0             # Fade-out duration at video end (seconds)
  fps: 30.0                     # Output frame rate
  mix_source_audio: false       # Mix original clip audio with the music track
  outro_padding: 2.0            # Padding after last clip before fade-out (seconds)
  source_volume: 1.5            # Volume multiplier for source audio when mixed
  target_res:                   # Output resolution [width, height]
    - 1920
    - 1080
  watermark_text: ""            # Watermark text overlay (leave empty for none)

generator:
  batch_size: 64                # Number of frames processed per CLIP batch
  clip_len_max: 5.0             # Maximum clip duration during ingest (seconds)
  clip_len_min: 2.0             # Minimum clip duration during ingest (seconds)
  clip_model: ViT-B-32          # CLIP model: ViT-B-32 (Standard) or ViT-L-14 (High Quality)
  clip_pretrained: laion2b_s34b_b79k  # CLIP pretrained weights (must match clip_model)
  sample_fps: 1.0               # Frame sampling rate during CLIP analysis
  scene_detection:
    embedding_threshold: 0.75   # Similarity below which consecutive frames are a hard cut
    drift_threshold: 0.92       # End a clip early once content drifts this far from its
                                # first frame. 0 disables.

privacy:
  discreet_default: false       # Start with Discreet Mode enabled
  pin_hash: ""                  # SHA-256 of the startup PIN (managed via the GUI)

paths:
  music_dir: ./music            # Default directory for music files
  output_root: ./output_clips   # Default directory for rendered output

jamendo:
  client_id: ""                 # Your Jamendo API key (free at developer.jamendo.com)

Editor Settings

`audio_crossfade`

Crossfade between consecutive tracks, used only when you supply more than one - separate several paths with ; or , in Audio Input. It has no effect on a single track, and nothing to do with your clips’ own audio. In the app this is Crossfade Between Tracks, beside the Audio Input field. Set to 0 for hard cuts between songs.

`fps`

Output frame rate. Use 30.0 for standard quality or 60.0 for smooth motion. Higher FPS doubles render time and file size.

`target_res`

Output resolution as [width, height]. Common values:

[1920, 1080] - 1080p (default)
[3840, 2160] - 4K
[1280, 720] - 720p (draft)

`mix_source_audio`

When true, the original audio from your source clips is mixed with the music track.

Both levels are constant for the whole render: the music holds at ducking_volume, your clips’ audio at source_volume. Earlier versions dropped the music only while they judged your clips’ audio to be “active”, which made the music pump up and down on continuous footage - that is gone.

source_volume - level of your clips’ own audio. 1.0 is as recorded.
ducking_volume - how loud the music sits underneath. 0.3 = 30%. Set it to 0 to mute the music entirely.

The mix is peak-limited, so raising either value cannot distort the render. In the app these are Source Volume and Music Level, under Source Audio Mixing on the Advanced tab.

The Mix Original Audio checkbox (Editor, GENERAL tab, under More options) overrides this key for a single render, in both directions: ticking it turns mixing on when the config file says false, and leaving it unticked turns it off when the config file says true.

⚡ NEW VIDEO has its own Keep your clips’ own sound under More options. It starts ticked when Mix Original Audio is ticked, uses the Editor’s Source Volume and Music Level, and the project remembers it, so the box matches when the project opens in the Editor.

This works in every render format, Triptych included. A Triptych has three panels on screen at once, so each clip is audible over its own cut rather than all three at the same time.

`watermark_text`

Text overlaid on the rendered output. Set to an empty string to disable. The Demo tier forces a watermark regardless of this setting.

Generator Settings

`batch_size`

Number of frames sent to the CLIP model per inference batch. Higher values use more VRAM but process faster. Reduce to 32 if you hit OOM errors during ingest.

`clip_len_min` / `clip_len_max`

Controls the minimum and maximum clip duration during scene detection. Shorter ranges produce more, shorter clips - better for fast-paced edits. Longer ranges preserve more continuous footage.

These set the length of the clips in your library, not how long each shot stays on screen. An edit cuts on the beat, so a shot shows only the part of a clip that fits between two cuts. To hold shots longer, set Pacing to Relaxed, Slow or Very slow in the EDIT section, or pick the SENSUAL style. These paces hold a shot for at most four bars of the music, so a slower song allows longer shots: about 7.5 seconds at 128 BPM, about 12 seconds at 80 BPM. Longest holds a shot for up to eight bars, ending on a bar line: up to about 15 seconds even at 128 BPM.

The Demo saves each clip as its own file, so a shot can’t run past the end of its clip: long shots need clips at least as long. Raise clip_len_min, or use Core or Studio, which read clips straight from your videos so shots can run as long as the music wants.

A scene shorter than clip_len_min gives no clip, so raising the minimum leaves out every shot in your videos that is shorter than it. Changes apply to videos analyzed afterwards: turn on Re-analyze videos already in the library when adding videos to apply them to your existing library.

`scene_detection.embedding_threshold`

Cosine similarity below which two consecutive sampled frames are treated as a hard cut. Lower values detect fewer cuts. Default 0.75.

`scene_detection.drift_threshold`

Ends a clip early once its content has drifted this far from the clip’s opening frame, so a single clip doesn’t span a change in the action. Only ever shortens a clip, so static footage keeps its normal length distribution while moving footage is cut on the change. Set 0 to disable. Default 0.92.

See Physical vs Virtual Ingest for what this does in practice.

`sample_fps`

How many frames per second are sampled for CLIP analysis. Higher values analyze more frames but increase ingest time, noticeably so without an NVIDIA GPU (about 2× longer at 4.0 on CPU). The default of 1.0 provides good coverage for most content.

Path Settings

`music_dir`

Default directory scanned for music files. Shown in the Editor’s AUDIO tab.

`output_root`

The projects folder: each project gets a folder here with its videos. Set it in ⚙️ Global Preferences → Projects folder (with Browse… and Open). Projects you’ve already made stay where they are when you change it. The chunks/ subdirectory is used for intermediate renders and can be safely deleted after rendering completes.

Jamendo Settings

`client_id`

Your Jamendo API key for free music discovery. Register at developer.jamendo.com (free, 2 minutes). Can also be set via the Jamendo panel or Global Preferences in the GUI.

Privacy Settings

`discreet_default`

When true, Onset Engine starts with Discreet Mode already enabled. Toggled from the GUI.

`pin_hash`

SHA-256 hash of the startup PIN. Empty means no PIN is set. Managed through the GUI - clear this value manually if you’ve forgotten the PIN.

Overriding via GUI

Most config values can be overridden in the Editor:

⚙️ Global Preferences modal for the projects folder, music dir, and watermark
Inspector panel for quality tier, FPS, and resolution
Per-project job files store project-specific overrides

Overriding these per project

Everything in config.yaml is a default. Nothing here is the last word - the settings you choose in the Editor win for that project, and they are saved with it, so two projects can disagree without you editing this file at all.

What you want to change	Where it lives in the Editor
Resolution and frame rate	Quality dropdown, bottom of the settings panel
Output format (16:9, triptych, vertical, highlights)	Format dropdown, bottom of the settings panel
Style preset	Style dropdown, Inspector
Pacing and narrative arc	Pacing (EDIT section) / Narrative (More options)
Custom driver	Driver → Browse, under More options
Restrict the clip pool	Source Collection, under More options
Keep source order	📅 Chronological Order, under More options
Steer clip choice by your ratings and choices	🧠 Learn my taste, under More options
Intro card	🔠 Title Card, under More options
Energy curve	Energy curve picker - see External Energy Curves
Output paths, music folder, watermark	⚙️ Preferences

Edit config.yaml when you want to change the default for every new project, or to reach a setting with no control in the interface - ingest sensitivity and scene-detection thresholds being the main ones.

Configuration Reference

Overview

Full Reference

Editor Settings

audio_crossfade

fps

target_res

mix_source_audio

watermark_text

Generator Settings

batch_size

clip_len_min / clip_len_max

scene_detection.embedding_threshold

scene_detection.drift_threshold

sample_fps