Version 0.2.3 · Pre-release
Practise with live posture and musical timing cues. Replay your drill with pivot and step overlays, then compare it with a teacher’s clip or an earlier take.
Experimental desktop build
TangoPosture is still in pre-release. Both installers include Python and the verified pose model. Camera, microphone and recording behaviour can vary by machine, and Windows support is experimental.
Desktop app features
Use a webcam for live coaching or analyse a saved clip. Review the movement, music and report together.
Live coaching
Track head alignment, raised shoulders, torso tilt and hip offset. A slow-tango cue flags repeated high ankle lifts. The coach estimates leader or follower arm patterns, open or closed embrace and knee flexion in closed embrace. Enable all-angle tracking for front and profile views, side steps, ochos and 360° pivots.
Calibration
Capture a baseline in the dedicated calibration camera with spoken prompts, a countdown and capture beeps, or select a stable interval in a video. Reopen a saved .tpc reference and restore its camera settings. Start reference-based scoring independently of recording.
Music and recording
The microphone estimates tempo and tracks eight-beat phrases, with manual Beat 1 alignment available. Record manually or opt into automatic start and stop when music is detected. A named .tp archive keeps numbered takes together, with microphone audio where supported.
Video and reports
Search a recording’s Markdown report and click a timestamp to seek. Compare raw footage, annotated video and a lines-only view using shared playback, speed controls and frame stepping. Optional audio playback uses ffplay or the system player.
Optional report analysis
Choose a model and an analysis focus. The app sends the report text to your configured Ollama server when you request analysis. Reopen saved responses in the practice archive or export them as Markdown.
Settings and help
Import or export TOML settings and save practice presets with your reference and file choices. Search the offline in-app manual or open a setting’s explanation directly. The command line also supports video analysis without a preview.
Movement replay
A hip-aligned ring appears during an estimated pivot and an arrow shows its direction. Recent heel-to-toe footprints fade out over a configurable window of five seconds by default. Saved practice runs are analysed automatically by default. You can also run or repeat the analysis in replay.
Reference clips
Import a local video up to 2 GB and analyse a segment up to 120 seconds, with optional reference-relative posture review. The desktop app also offers optional YouTube and Facebook Reel downloads via yt-dlp for footage you’re authorised to use. YouTube source links can open the matching moment after you align the timestamps.
Comparison
Play and seek through both clips side by side. Align a shared moment such as the start of a pivot. Toggle coach lines, pivot markers and footprints, or hide the footage behind a neutral floor guide to focus on movement.
Posture analytics
Compare practice and reference traces for head alignment, shoulder height, knee bend and hip alignment or offset. Zoom into a section, pan across it and copy a chart image.
Soundtrack
Optionally analyse the imported clip’s own soundtrack for estimated beat and phrase timing. The analysis runs locally using the video audio or a WAV sidecar.
Clips and export
Cut a selected range into a new clip or export a replay as an annotated MP4 with the available coach, pivot and step overlays you choose.
For pivot and step review, film one dancer with a fixed camera and keep both feet in frame. The website preview below uses an illustration. Video analysis runs in the desktop app.
01 / Practice workflow
Start with a short solo drill. Capture a baseline, then return to the same moment in your video and its movement report.
Capture a reference stance with a steady camera. Save the calibration so you can review it or reuse it with the same setup.
.tpc · calibrationStart and stop recording manually for your first take. The app can save raw footage alongside the annotated overlay and report.
.tp · practice archiveJump to a report timestamp, compare the raw and annotated views, and bring a specific observation to your next lesson.
Video + report02 / Overlay preview
Try the estimated pivot and recent-step layers, or switch to the original pose views. Each option shows one sample replay frame.
During an estimated pivot, a ring follows the hip line. The arrow shows the estimated direction of rotation.
03 / Community pilot
TangoPosture is a free, non-commercial open-source project. Dancers and teachers can try the 0.2.3 pre-release on Apple Silicon Macs and Windows x64 PCs and share feedback on their setup and practice workflow.
GitHub issues are public. Share your OS, the exercise and a description of what happened. Keep personal recordings and reports private.
The app estimates visible landmarks and compares selected observations with a reference. Camera angle, lighting, clothing, turns and a partner obscuring the view can reduce coverage. Use the footage and your teacher’s feedback to interpret an observation. The software does not provide a medical assessment or a validated judgement of tango technique.
Core capture and pose analysis run locally after setup. Practice archives can include video, microphone audio and reports. Automatic song recording is opt-in and can also capture speech or bystanders. If you explicitly request optional analysis with a remote Ollama server, the app sends report text to that server.
The public repository contains the application source and development guides. The application code is released under Apache 2.0. Dependencies and model weights have separate terms.
TangoPosture