Click a shot. Scrub a frame. See what the joints actually did.
No install, no login. This is a replay of a recorded run.
Change three moves, then return to the model’s home pose. Download a plan and render it locally. The recorded video above stays unchanged.
These controls edit a JSON plan, not a live simulator. Local runs label these motions “scripted.”
robot-reel --shots shots.json --output artifacts/my-film
Python 3.12+ and OpenGL. First run downloads robot assets. No model credentials for the scripted demo.
git clone https://github.com/noteflowai/robot-reel.git cd robot-reel python3.12 -m venv .venv source .venv/bin/activate pip install -e . robot-reel --output artifacts/demo python -m robot_reel.verify artifacts/demo
Each new run includes this replay page beside the videos. Open artifacts/demo/index.html directly in your browser.
The SO-100 uses position actuators and MuJoCo physics steps. The G1 segment uses scripted joint-space poses with a fixed root; it is not a walking or balancing policy. Arm targets show the destination for each shot. Microduck targets show the policy actuator command. Driving references show nominal early-braking standstill values.
The replay displays the supplied recording. The command-line verifier checks file consistency and trace agreement; hashes are not signatures and do not establish independent authenticity or autonomous task success.