Diffusers Web

This project is a web based image generation tool, a front-end wrapper for HuggingFace's diffusers.

Usage

Run the web server:

python ./main.py

If this works, navigate to http://localhost:8157 in your browser.

Features

Text to image

Image transform

Navigate to http://localhost:8157/converter.

Inpainting and outpainting

Before run web server, config the diffuser model to an inpainting model. E.g. add an environment variable in .env.local:

DIFFUSER_MODEL_PATH=runwayml/stable-diffusion-inpainting

Then navigate to http://localhost:8157/painter.

Tips: image copy in a web page other than localhost requires https protocol, config SSL_CONTEXT="'adhoc'" in .env.local to achieve this.

Requirements

Hardware

A GPU with CUDA capability is required for good performence. The more video memory the better, 24GB is recommended.

Python Libraries

Install dependency libraries with pip, reference to requirements.txt.

Authorization

You may need a HuggingFace access token to download pretrained models for the first running.

Local Environment Variables

Edit local environment variables in file .env.local.

Variable Name	Default Value	Description
HF_TOKEN		Your HuggingFace access token. If a local config path provided, this can be ignored.
DIFFUSER_MODEL_PATH	stabilityai/stable-diffusion-2	This can be a local model config path.
TEXTGEN_MODEL_PATH	k-l-lambda/clip-text-generator	The random painting description generator model path. This can be a local model config path.
HTTP_HOST	127.0.0.1	Use 0.0.0.0 for network access.
HTTP_PORT	8157
SSL_CONTEXT	None	Use 'adhoc' for https server.
DEVICE		cuda or None
TEXT_DEVICE_INDEX	0	This can be greater than 0 if CUDA_VISIBLE_DEVICES has more than 1 gpu specified.