azailib

AI models like Stable Diffusion require image pre-process, generate canny image etc. this repo is for holding these common tool in one repo.

Install

Install azailib:

pip install -U git+https://github.com/xhinker/azailib.git@main

Then install required packages from repo source:

pip install -U git+https://github.com/xhinker/sd_embed.git@main
pip install -U git+https://github.com/facebookresearch/sam2.git@main

Note that the sam2 installation require package building, will take some time, be a bit patient, read some news or drink a cup of coffee.

If see any message says onnxruntime-gpu is needed, install it using this command:

pip install onnxruntime==1.16.2 --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-11/pypi/simple/

Use it

A sample to use azailib to load and use 8-bit flux.1 pipe

import torch
from azailib.sd_pipe_loaders import load_flux1_8bit_pipe

model_path  = "/home/andrewzhu/storage_14t_5/ai_models_all/sd_hf_models/black-forest-labs/FLUX.1-dev_main"
pipe        = load_flux1_8bit_pipe(
    checkpoint_path_or_id   = model_path
    , pipe_device           = "cuda:0"
)

#%%
# generate image
prompt = "4k, best quality, beautiful 20 years girl, show hands and fingers"
image = pipe(
    prompt                  = prompt
    , guidance_scale        = 3.5 
    , num_inference_steps   = 24
    , height                = 1024
    , width                 = 1024
    , generator             = torch.Generator('cuda:0').manual_seed(5)
).images[0]

display(image)

That is it.