VELA product-loop kit

Unzip the whole folder, then open index.html in a browser. Videos use relative paths and work offline.

source.png is an AI-generated fictional VELA citrus tonic still made with Nano Banana 2, not a photograph of an actual commercial product.

pixelcut.mp4: first Pixelcut Looping Video take.
matched-endpoints.mp4: MiniMax H3 Max take using source.png as first and last image.
pixelcut-v2.mp4: revised Pixelcut take, normalized by the Voyager agent to 1920x1080, five seconds, silent.
product-loop.mp4: frames 7–116 inclusive of v2,110frames at24fps,4.583seconds.

JSON files contain exact submitted generation prompts/parameters with local source references. Running these again incurs model charges and may produce different output. Use Voyager models run --input @filename.json --json from this folder, or adapt native input objects to your provider.

The source image and generated footage are provided for learning and adaptation. The generation route does not guarantee exact product-pixel preservation.

Reproduce the trim:
ffmpeg -i pixelcut-v2.mp4 -vf "trim=start_frame=7:end_frame=117,setpts=PTS-STARTPTS" -an -c:v libx264 -crf 18 -pix_fmt yuv420p -movflags +faststart product-loop-new.mp4

Finished label composite:
product-loop-final.mp4 restores the source lettering over the generated label.
product-loop-final-three-repeats.mp4 is the final three-cycle review clip.

Reproduce label composite: install Python3, numpy, opencv-python-headless, and FFmpeg.
python3 -m pip install numpy opencv-python-headless
Move the supplied product-loop-final.mp4 elsewhere first (the script refuses overwrites).
python3 label-composite/composite_label.py

The bundled alignment and mask are specific to this photo and fixed camera. A different shot needs new alignment. The script fits the blank paper color around the text and feathers its mask; it does not use generation.

The downloadable videos are compressed H.264 copies; private lossless intermediates are not included.
