Pass-O-Guava / StableDiffusion-demo

Stable Diffusion Python Demo

1. Install

I. Create Project and downlod weights.

mkdir stablediffusion-demo
mkdir sd_models
cd sd_models

# stable-diffusion-v1-5 (66G)
sudo apt-get install git-lfs
git init
git lfs install
git clone https://huggingface.co/runwayml/stable-diffusion-v1-5

# stable-diffusion-v1-5-inference (5.2G, recommend!)
链接: https://pan.baidu.com/s/1_RS94vJIdbwpmPbhRKmQnw?pwd=yzne 
提取码: yzne

conda activate stablediffusion
pip install torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install transformers==4.19.2 diffusers==0.12.1 invisible-watermark

II. Files tree of "stable-diffusion-v1-5":

stable-diffusion-v1-5/
|-- [ 14K]  README.md
|-- [4.0K]  feature_extractor
|   `-- [ 342]  preprocessor_config.json
|-- [ 543]  model_index.json
|-- [4.0K]  safety_checker
|   |-- [4.6K]  config.j
|   |-- [1.1G]  model.safetensors
|   `-- [1.1G]  pytorch_model.bin
|-- [4.0K]  scheduler
|   `-- [ 308]  scheduler_config.json
|-- [4.0K]  text_encoder
|   |-- [ 617]  config.json
|   |-- [469M]  model.safetensors
|   `-- [469M]  pytorch_model.bin
|-- [4.0K]  tokenizer
|   |-- [512K]  merges.txt
|   |-- [ 472]  special_tokens_map.json
|   |-- [ 806]  tokenizer_config.json
|   `-- [1.0M]  vocab.json
|-- [4.0K]  unet
|   |-- [ 743]  config.json
|   |-- [3.2G]  diffusion_pytorch_model.bin
|   `-- [3.2G]  diffusion_pytorch_model.safetensors
|-- [4.0G]  v1-5-pruned-emaonly.ckpt
|-- [4.0G]  v1-5-pruned-emaonly.safetensors
|-- [7.2G]  v1-5-pruned.ckpt
|-- [7.2G]  v1-5-pruned.safetensors
|-- [1.8K]  v1-inference.yaml
`-- [4.0K]  vae
    |-- [ 547]  config.json
    |-- [319M]  diffusion_pytorch_model.bin
    `-- [319M]  diffusion_pytorch_model.safetensors

Select inference model files and config files.

stable-diffusion-v1-5-inference
├── [4.0K]  feature_extractor
│   └── [ 342]  preprocessor_config.json
├── [ 543]  model_index.json
├── [4.0K]  safety_checker
│   ├── [4.6K]  config.json
│   └── [1.1G]  pytorch_model.bin
├── [4.0K]  scheduler
│   └── [ 308]  scheduler_config.json
├── [4.0K]  text_encoder
│   ├── [ 617]  config.json
│   └── [469M]  pytorch_model.bin
├── [4.0K]  tokenizer
│   ├── [512K]  merges.txt
│   ├── [ 472]  special_tokens_map.json
│   ├── [ 806]  tokenizer_config.json
│   └── [1.0M]  vocab.json
├── [4.0K]  unet
│   ├── [ 743]  config.json
│   └── [3.2G]  diffusion_pytorch_model.bin
└── [4.0K]  vae
    ├── [ 547]  config.json
    └── [319M]  diffusion_pytorch_model.bin

III. Inference Models:

Model	Name	File Size
text_encoder	pytorch_model.bin	470M
unet	diffusion_pytorch_model.bin	3.3G
vae	diffusion_pytorch_model.bin	320M
safety_checker	pytorch_model.bin	1.2G
Total: 5.2G

2. Test:

I. txt2img

python txt2img.py

(stablediffusion) ubuntu@wilson:~/wy/AIGC/stablediffusion-demo$ python txt2img.py 
WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
    PyTorch 1.12.1+cu113 with CUDA 1105 (you have 1.12.1+cu113)
    Python  3.9.16 (you have 3.9.16)
  Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
  Memory-efficient attention, SwiGLU, sparse and more won't be available.
  Set XFORMERS_MORE_DETAILS=1 for more details
A matching Triton is not available, some optimizations will not be enabled.
Error caught was: No module named 'triton'
100%|███████████████████████████████████████████████████████████████████████████████████| 50/50 [00:17<00:00,  2.92it/s]
==> nsfw_content_detected: False
==> image saved.

II. img2img

python img2img.py

Plan

• txt2img
• img2img
• inpaint
• others scheduler(default PLMS/DDIM/K-LMS/CycleDiffusion)
• others SD model

Acknowledge

https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines/stable_diffusion