AI-WEBUI: A universal web interface for AI creation, a handy tool for image, audio, and video processing

⭐ If it helps you, please give it a star, thank you! 🤗 中文文档

🌟 1. Introduction

ai-webui is a browser-based interface designed to provide a universal AI creation platform.

This project provides basic functionalities such as image segmentation, object tracking, image restoration, speech recognition, speech synthesis, as well as advanced features such as chatbot, video translation, and video watermark removal, which greatly improve the efficiency of short video creation.

⚡2. Installation

To install and use AI-WebUI, follow these steps:

2.1 Clone this project to your local machine

git clone https://github.com/jasonaidm/ai_webui.git

2.2 Enter the project directory

cd ai_webui

2.3 Create a virtual environment

conda create -n aiwebui python=3.11
conda activate aiwebui

2.4 Install the required dependencies

apt install ffmpeg -y 
pip install -r requirements.txt

🚀3. Quick Start

Using AI-WebUI is very simple. Just follow the instructions on the interface. You can input creative elements by uploading videos, audios, images, or entering text, and interact with the model's output.

python webui.py -c ./configs/webui_configs.yml

After starting, open http://localhost:9090/?__theme=dark in your browser to see the project interface.

3.1 Single Function Demo

Considering the GPU performance issues of some users' personal computers, we provide single function demos that allow users to run a specific AI function without starting the entire project.

Image Segmentation

Panorama segmentation
Segmentation based on points coordinates
Segmentation based on textual prompts

python webui.py -c ./configs/segmentation_demo.yml

Speech Recognition

Multilingual speech recognition (e.g., Chinese and English)

python webui.py -c ./configs/asr_demo.yml

Speech Synthesis

Multilingual speech synthesis (e.g., Chinese and English)

python webui.py -c ./configs/tts_demo.yml

3.2 Combined Function Demo

More complex functions can be obtained by combining multiple AI models, requiring higher GPU resources.

Chatbot

Text-based chatbot
Voice-based chatbot

python webui.py -c ./configs/chatbot_demo.yml

Video Restoration

Watermark removal
Mosaic removal
Object tracking
Object removal in videos

python webui.py -c ./configs/video_inpainter_demo.yml

Video Conversion

Audio-video separation
Image cropping
Image noise addition
Frame extraction
Speech recognition
Subtitle translation
Speech synthesis
BGM addition
One-click video generation (automatic video replication from the internet)

python webui.py -c ./configs/video_convertor_demo.yml

3.3 Full-function Online

Open all AI functions by running the following command:

python webui.py -c ./configs/webui_configs.yml

Since model loading takes a long time, it is recommended to load the models during the first inference after starting. You can control the loading strategy of each AI model through the "init_model_when_start_server" option in the configs/base.yml configuration file.

🔥4. Model Files

4.1 Model File Downloads

Model	Model File Size	Small Model List	Download Link
chatglm2-6b-int4	3.7G	✅	Baidu Netdisk
chatglm2-6b	12G		Tsinghua University Cloud Disk
sam_vit_b	358M	✅	Baidu Netdisk
sam_vit_h	2.4G		Baidu Netdisk
FastSAM-s	23M	✅	Baidu Netdisk
FastSAM-x	138M		Baidu Netdisk
ProPainter	150M	✅	Baidu Netdisk
raft-things	20M	✅	Baidu Netdisk
recurrent_flow_completion	19M	✅	Baidu Netdisk
cutie	134M	✅	Baidu Netdisk
whisper-samll	461M	✅	Baidu Netdisk
whisper-large-v3	2.9G		Baidu Netdisk

The extraction code for Baidu Netdisk is: zogk

4.2 Directory Structure of Model Weight Files

model_weights/
├── chatglm
│   └── chatglm2-6b-int4
│       ├── config.json
│       ├── configuration_chatglm.py
│       ├── modeling_chatglm.py
│       ├── pytorch_model.bin
│       ├── quantization.py
│       ├── tokenization_chatglm.py
│       ├── tokenizer.model
│       └── tokenizer_config.json
├── fastsam
│   ├── FastSAM-s.pt
│   └── FastSAM-x.pt
├── propainter
│   ├── ProPainter.pth
│   ├── cutie-base-mega.pth
│   ├── raft-things.pth
│   └── recurrent_flow_completion.pth
├── sam
│   ├── sam_vit_b.pth
│   └── sam_vit_h.pth
└── whisper
    ├── large-v3.pt
    └── small.pt

If the GPU memory is less than 8G, you may need to use the small models to run the project; however, the performance of the small models may not be ideal, so it is recommended to run the large models if possible.

5. Contributing

If you have any suggestions or feature requests, please feel free to create an issue.

6. References

Segment-ant-Track-Anything
ProPainter
ChatGLM2-6B
segment-anything
FastSAM
whisper

lh08164311/ai_webui