Give Motion Lab a reference clip. Your character performs the same motion — rendered 100% locally on your Mac. No per-second credits. No uploads. No watermarks.
A 5.0-second reference performance (121 frames) was reduced to a DWPose skeleton, then replayed by a completely different character with LTX-2.3 MLX + IC-LoRA — 100% on a Mac, offline. Real output only, QA numbers included below.
Lab note, Oct 24 2026: Prototype #1 holds the seated posture and the head-turn sequence with a consistent face and outfit, no morphing. Known flaw: the camera angle drifts from the reference (fix already in production on take #2). We publish the flaw with the demo — that is the deal.
Cloud video bills you for every second, every retry. Motion Lab runs on hardware you already own.
Cloud plans charge per second and monthly credits expire. Local renders cost electricity — iterate as many times as the shot needs.
Reference footage and characters never leave your Mac. Suitable for client work and unreleased material.
Pose and depth are extracted from your reference (DWPose + IC-LoRA), so the performance is transferred — not guessed from text.
Three steps, one pipeline: ToonFlow for story, ComfyUI/MLX for production, Motion Lab for motion.
Drop a short clip with the action you want. Scenes are auto-detected and keyframes extracted.
A vision model writes a per-shot spec (character, action, camera, light). DWPose extracts the body, hand and face motion.
LTX-2.3 on MLX replays the motion with your character and scene. Every output passes a dense-frame QA gate — no looping actions, no warped frames.
Reference benchmarks from the open MLX stack we build on, plus our own measured M3 Max numbers from Demo #1 above.
| Configuration | Machine | Clip | Render time |
|---|---|---|---|
| LTX-2.3 q4 distilled (MLX) | M4 Max, 64GB | 2s · 49 frames | 88s (+34s load) |
| Wan 2.2 5B (MLX, for comparison) | M3 Max, 64GB | 1.7s · 41 frames | 17m 43s |
Sources: upstream MLX benchmarks documented in our Oct 23, 2026 research report. We do not publish user counts, revenue or testimonials we do not have.
The tool is in beta. The service is open now.
Send a reference clip + character image. We return the remake, QA-passed. 15s $79 · 30–45s $149 · 90s $299.
ComfyUI/MLX workflows + motion presets + install guide. Starter $49 · Pro $99. Join the waitlist.
One-click local app (beta): LTX-2.3 MLX + DWPose + IC-LoRA + storyboard bridge. Lifetime beta, or Pro $29/mo.
System requirements: Apple Silicon M1–M4 · 32GB RAM minimum (64GB recommended) · 60–120GB free disk · macOS 14+.