tahercoolguy / video_background_remover_appender

Remove Background of video and add yours

  • Public
  • 78 runs
Run with an API
Playground API Examples README Versions

Examples

View more

Run time and cost

This model runs on Nvidia A40 (Large) GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

Background Remover Cog Model for Replicate

This repository contains a Cog model for running the Background Remover video processing model on Replicate. The Background Remover is an AI model that removes or replaces the background in videos.

How it Works

The model uses a two-step process:

  1. Background Segmentation: Uses the BiRefNet model to create a mask for each frame, separating the foreground from the background.
  2. Background Replacement: Replaces the background with either a solid color, an image, or another video.

Running on Replicate

This model is designed to run on Replicate only.

Inputs:

  • input_video: The input video you want to process.
  • bg_type: The type of background to use (Color, Image, or Video).
  • bg_image: The background image to use (if bg_type is Image).
  • bg_video: The background video to use (if bg_type is Video).
  • color: The background color to use (if bg_type is Color).
  • fps: Output FPS (0 will inherit the original fps value).
  • video_handling: How to handle background video if it’s shorter than the input video (slow_down or loop).

Outputs:

  • A video in MP4 format with the background removed or replaced.

How to Use

  1. Upload your input video.
  2. Choose the background type (Color, Image, or Video).
  3. Depending on your choice, provide the necessary background (color, image, or video).
  4. Adjust the FPS if needed (0 to keep original).
  5. If using a background video, choose how to handle it if it’s shorter than the input video.
  6. Run the model and wait for the processed video.

Technical Details

The model uses the following key components:

  • BiRefNet for image segmentation
  • PyTorch for deep learning operations
  • MoviePy for video processing
  • Pillow for image processing

The main processing steps are:

  1. Load and prepare the input video
  2. Process each frame:
  3. Apply the BiRefNet model to create a mask
  4. Composite the frame with the chosen background
  5. Reassemble the processed frames into a video
  6. Add the original audio back to the processed video

Running Locally (Optional)

To run this model locally, you’ll need:

  • Python 3.10
  • CUDA-compatible GPU
  • Cog (pip install cog)

Follow these steps:

  1. Clone this repository
  2. Navigate to the project directory
  3. Run cog predict -i input_video=@path_to_your_video.mp4 -i bg_type="Color" -i color="#00FF00"

Replace the input parameters as needed.

Credits

  • This model uses the BiRefNet architecture for image segmentation, developed by ZhengPeng7.
  • Special thanks to multiplewords.com for providing the model on Replicate.