Web application for cover generation, based on the provided video, description, and optional parameters.
It can be handy for people without design experience or for those who wish to get a cool cover as soon as possible.
The resulting cover consists of:
- Speaker
- Text
- Background
Please, take a look at our presentation for more details and examples.
The user is asked to provide the following parameters:
-
Video
-
Description
It is used for text and cover background generation.
-
Background style:
- Generate background using a diffusion model
- Create a background from one or two stitched video frames
-
Text color
-
User photo [optional]
Cut the user from the provided photo rather than selecting a random person from the video.
- Web application
- Cover generation module
- Person detection - RetinaFace
- Human image matting (segmentation) - MODNet
- Text summarization for keywords - PyMorphy2
- Text summarization for a clickbait phrase - T5-Russian
- Background generation - Kandinsky 2.2
For running web service, run the following command from the project directory:
python3 manage.py runserver [port]
A port can be, for example, 8080. The web server can be accessed on:
http://127.0.0.1:8080/generate
NOTE: the used models are heavy, so two video cards 32 GB are needed.
🐍 Python [Django]
⏳ Javascript [AJAX]