mcmonkeyprojects / modelscope-text2video-webui

Simple Locally-Ran ModelScope Text2Video WebUI

ModelScope Text2Video WebUI

Dirt simple Gradio powered WebUI for https://modelscope.cn/models/damo/text-to-video-synthesis/summary

Install And Basic Usage

• Install Python 3.7 (or 3.10 or whatever idk it's probably fine)
• Run the program
  • For Windows, double-click start.bat
    • If something breaks in install, delete venv and run start.bat again
  • For Linux, run ./start.sh
    • If something breaks in install, run ./install.sh
• Wait a while, it installs a ton of python packages then downloads like 20 gigs worth of random models
  • You might need to restart it a few times, it will randomly crash because they have server issues I guess.
• Open http://localhost:7860/ in your browser.
• Type prompts, hit generate, enjoy.

Advanced Usage

• Want to change the port or whatever? Edit it at the bottom of the modelscope_gradio.py file.
  • I'll add better options for all that if people start actively using this repo, for now this is a day-one slap together interface to make it easier than the mess the original repo left users with.

Notes

• Some files wind up in /tmp
• Some model files for some reason wind up in your home directory at ~/.cache/modelscope/
• Outputs will be stored in ./outputs/
• Takes about 20 seconds to generate a video on an RTX 3090
• Uses about 16 GiB of VRAM

Licenses

Code in this repo is MIT (do whatever ya want with it), modelscope is Apache2, the model itself has its own license. Other licenses from Gradio and other upstream python libs apply however relevant.