Free PMV driver pack - 5 tuned drivers, drop in and render

Drivers are Onset Engine's semantic layer: plain-JSON files that tell the AI which clips belong at each energy level of the song. The engine ships content-neutral - so here are 5 drivers tuned specifically for the edits this site is about, with the wordings measured rather than guessed. Free, no sign-up.

⬇ Download the driver pack (5 KB)

What's in the pack

• PMV Classic - Tease to Peak. The canonical arc, escalating by act: tease on quiet sections, foreplay through the verses, sex on the build, climax on the drops. Pair with the Aggressive or Sensual preset.
• HMV. Tuned for animated source footage, escalating by act. Honest caveat: I tune these against my own live-action library, so this one's tier structure is verified but its retrieval quality isn't — if it misses on your footage, the testing command below will tell you which tier to reword. Pair with the AMV preset.
• Cock Hero - Round Escalation. Beat-game structure: rounds escalate by act, skewed tier thresholds keep them distinct instead of flickering, and the peak tier strongly favours your 4-5 star clips. Pair with long-holds pacing and the beat grid export for your beat bar overlay.
• Hypno Loop. Trance repetition: close-up heavy and symmetrical, with tiers that deliberately share clips so the edit keeps circling a similar look. The one driver here where overlap is the goal. Pair with the Hypnosis preset at half-beat pacing.
• Focus: Rear. A single-subject driver that escalates by what is in frame — clothed, then underwear, then explicit. Use it as the template for focusing on anything else (see below).

Install (30 seconds)

1. Unzip the pack anywhere you like — Documents is fine. There's no special folder, and you don't need admin rights
2. In Onset Engine, under Creative Direction → Driver, click Browse and pick one of the .json files
3. Add a preset and your track, then render. No restart needed

Onset Engine loads a driver by its full path, so the file can live wherever suits you. The drivers bundled with the app sit in C:\Program Files\Onset Engine\drivers\examples if you want to read them for reference — but keep your own drivers somewhere writable instead, so a reinstall can't wipe them.

Writing a driver that actually works

Driver quality comes down almost entirely to how you word the descriptions. I tested this properly — scoring candidate wordings against a real 1,200-clip library and measuring which ones retrieved a distinct set of clips. Two findings changed how I write them:

1. Write tags, not sentences

The AI learned from image captions and alt-text, so it responds to how people actually label pictures — short, blunt, concrete. Longer descriptive prose scores worse, not better. Same concept, measured:

• ass — strongest match of everything tested
• ass in thong — nearly as strong, and far more specific
• rear view of a woman bending over — noticeably weaker
• close-up of a woman's buttocks — weakest of the lot

Use the words you'd actually type into a search box. Clinical vocabulary and full sentences both hurt.

2. Escalate by what's in frame, not by speed

This is the big one. If your tiers describe the same thing at different speeds, they'll play the same clips — the AI reads content far better than it reads pace.

The Focus driver originally escalated by pace ("gentle unhurried movement" → "blurred fast motion"). Its tiers shared up to 27% of their clips. Rewriting it to escalate by what's actually visible — clothed, then underwear, then explicit — dropped that to 4–13% across almost every pair. Same subject, same library, completely different result.

The same fix has since been applied to the rest of the pack. Cock Hero's tiers were originally four flavours of "fast rhythmic movement" and shared 19–30% of their clips on every pair; rebuilt as an act ladder they now share 0–18%. PMV Classic went from a worst pair of 25% to 19%.

So build your ladder out of things you could point at in a still frame: clothing state, framing, act, toys. Not adverbs.

If you run Onset Engine from source, you can measure your own wording before committing to it:

python tools/test_prompts.py --driver mydriver.json

It reports whether each prompt finds a distinct pool, and whether your tiers genuinely pull different clips from one another.

Make them yours

The descriptions inside each driver are plain-text prompts. Open any of them in a text editor and reword the tiers for your specific library - it's just JSON. Two upgrades worth making:

• Rate your clips. The peak tiers prefer 4-5 star clips, so your best material lands on the drops. It's a strong preference rather than a hard rule — an unrated clip can still be picked if nothing better fits the moment, so you're never left with a gap.
• Tag your performers in the Subjects tab, then add "tags": ["@Name"] to any tier to steer it toward a specific performer. It's a strong pull rather than a lock — the tag counts for 30% of the score by default, so a perfect content match can still win. Raise "tag_weight" in the driver's global block if you want the performer to dominate.

For the full schema — every field, what the mood and scene-type preferences actually do to a clip's score, and how the tier thresholds map energy to tiers — see the official JSON Driver Masterclass in the Onset Engine docs.