CoolFace
Apppublic

dimiporf/parcel-conveyor-video

sourceHugging Faceupdated 20d agoView on Hugging Face
0likes
App README

Parcel Conveyor — continuous AI video

This Hugging Face Space produces a single continuous 10- or 30-second parcel-conveyor video using the open SkyReels V2 Diffusion Forcing 1.3B model. It runs in image-to-video mode so the uploaded first frame fixes the camera position, conveyor layout, and warehouse appearance.

Deploy on the free tier

  1. 1.Create a Hugging Face account and verify its email. A free account in good standing and at least 30 days old can host up to two ZeroGPU Spaces.
  2. 2.Create a new Gradio Space, then select ZeroGPU hardware in its settings.
  3. 3.Copy the three files in this directory to the Space repository and commit them.
  4. 4.Wait for the first build and model download. Upload a clean, 16:9 reference image of the conveyor and select 30 seconds.

The free ZeroGPU tier has a daily quota and a shared queue. Keep this Space private if the reference frame contains confidential information; verify the account and Space visibility policy before uploading company material.

Recommended reference frame

Use a 16:9 still with a fixed overhead or oblique CCTV camera, empty conveyor or a small number of unbranded parcels, and visible side rails. Avoid readable labels, people, moving machinery outside the belt, and strong perspective changes.

What the controls do

  • —30 seconds uses the official SkyReels long-video setting of 737 frames at 24 fps.
  • —Seed lets you repeat an exact generation when all other settings remain the same.
  • —Inference steps: start at 20; increase only after a promising test, because higher values consume more shared GPU time.

Generated footage is suitable for a visual demo. It must not be used as labelled detector validation data: generated parcels can change shape, count, and position from frame to frame.

Sources