Everything the engine actually does.

No marketing fog. This is the literal feature list of the RHYTHMIX Studio CLI engine that ships today.

v0.3.0 · Released today

Audio pipelineFrom your track to a scene plan

🎧

Audio probing

Reads duration, sample rate, channel count and codec via ffprobe. Supports MP3, WAV, M4A, FLAC and anything ffmpeg knows.

📊

Loudness-driven structure

Samples per-second RMS loudness across the whole track and groups loud/quiet runs into intro / verse / chorus / bridge / outro. Loud passages become chorus, quiet passages become verse.

--flat-plan to disable
🥁

Beat-snapped cuts

Section and scene boundaries snap to the nearest beat. Pass --bpm or install aubio for auto-detect.

--bpm 120 / brew install aubio
📝

Scene-aware prompt library

5+ prompts per role (intro/verse/chorus/bridge/outro), deterministically shuffled so consecutive scenes in the same section never repeat a prompt.

Generation sourcesThree modes, one engine

ModeCostVisual qualityRequires
Replicate (paid) ~$8–25 / render AI-generated, theme-matched REPLICATE_API_TOKEN
Pexels (free) $0 Real stock footage PEXELS_API_KEY (free signup)
Local (free) $0 Whatever clips you supply Folder of .mp4 files

Models on Replicate4 top video models, auto-routed

ModelProviderBest forMax clip
Kling v2KuaishouCinematic chorus shots10s
Hunyuan VideoTencentMotion-heavy verses5s
Luma RayLuma LabsDreamy bridges, smooth camera5s
MiniMax HailuoMiniMaxCharacter / expressive6s

The planner picks a model per section automatically. Override with --model <name>.

Compose pipelineFrom scenes to MP4

📐

Three aspect ratios

16:9 landscape (1280×720), 9:16 portrait (720×1280) for TikTok/Reels, or 1:1 square (1024×1024) for Insta feed. Auto-crop fills the frame regardless of source clip dimensions.

Crossfade transitions

0.25s xfade between every scene with mathematical duration compensation so the final video matches your audio length exactly. --no-transitions for hard cuts.

♻️

Retry & checkpoint

Each scene fetch retries up to 3× with exponential backoff. Finished scenes are kept on disk; a failed render resumes from the last successful scene instead of re-spending.

💰

Dry-run cost preview

See the per-scene model breakdown and estimated cost before any API call fires. Edit plan.json to tweak prompts or models, then render-from-plan.

EngineeringBuilt to not surprise you

🧪

Integration tests

27 tests covering loudness curve, structure detection, all 3 aspect ratios, crossfade + cut modes. npm test runs the full pipeline on a synthetic track.

📦

Zero npm dependencies

Pure Node 20 + ffmpeg. The whole zip is 20 KB. No node_modules, no supply-chain risk, no version drift.

🔄

Replicate 429 backoff

Honors retry_after from rate-limit responses with jitter. Long renders won't get throttled out.

On the roadmapComing next — included in your $149

Next 14 days

Studio web app

Browser upload + render queue. No Node, no terminal, no env vars. Same engine underneath.

Q3 2026

LLM-driven prompts

Per-scene prompts generated from theme + section + song mood instead of template recipes.

Q3 2026

Beat-locked transitions

Whip-pan / dip-to-color transitions that fire on downbeats in chorus.

Q4 2026

Onset-detected structure

Section boundaries from spectral novelty, not just loudness — picks up texture changes loudness misses.

$149 once. Every future feature included.

30-day money-back guarantee. No subscription.

Buy RHYTHMIX — $149 →