https://github.com/riccardobl/stable-diffusion-workspace-docker
https://github.com/riccardobl/stable-diffusion-workspace-docker
Last synced: about 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/riccardobl/stable-diffusion-workspace-docker
- Owner: riccardobl
- Created: 2022-09-22T09:56:23.000Z (over 2 years ago)
- Default Branch: master
- Last Pushed: 2022-10-30T11:24:26.000Z (over 2 years ago)
- Last Synced: 2024-10-06T05:21:14.159Z (8 months ago)
- Language: Python
- Size: 47.9 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Stable Diffusion Workspace
A docker container preloaded with stable diffusion and cuda drivers.
I've built this for my personal, don't expect it to be properly documented or supported.
## Requirements
- Linux Host
- Docker
- NVIDIA gpus with proprietary drivers and cuda
- [Huggingface](https://huggingface.co/)'s username and apikey to download the model (free)
- At least 16 GB for ram
- A decent nvidia GPU with >=8 GB vram
## Usage / Cheatsheet
### Run workspace
```bash
docker run
-eHUGGINGFACE_USER="%%USERNAME%%" \
-eHUGGINGFACE_PASSWORD="%%API_KEY%%" \
-d \
-v /srv/stable-diffusion:/data \
--name="stable-diffusion-workspace" \
--gpus all --shm-size="8g" --ulimit memlock=-1 --ulimit stack=67108864 \
riccardoblb/stable-diffusion-workspace
```
Note:
- Decrease shm-size if your machine doesn't have enough ram (a shm-size of 1gb should be enough, probably)
### Open a shell inside the workspace
```bash
docker exec -it stable-diffusion-workspace bash
```
### Download a pinterest collection as dataset
```bash
download_dataset.sh user/collection
```
### Fine tuning
Tweak the yaml files `~/InvokeAI/configs/stable-diffusion/v1-finetune_style.yaml` or `~/InvokeAI/configs/stable-diffusion/v1-finetune.yaml` to your liking.
See: https://towardsdatascience.com/how-to-fine-tune-stable-diffusion-using-textual-inversion-b995d7ecc095
and: https://www.youtube.com/watch?v=WsDykBTjo20
to get a better understanding on what to do.
Then run
```bash
cd ~/InvokeAI
python3 main.py --base ./configs/stable-diffusion/v1-finetune_style.yaml -t --actual_resume ~/InvokeAI/models/ldm/stable-diffusion-v1/model.ckpt -n SOMETHING --data_root /data/datasets/YOUR_DATASET --precision 32 --gpus 0,
```
( --precision 32 is for gpus that don't support half precision (1xxx gpus and <))
### Interactive Terminal
```bash
python3 ./scripts/dream.py
```
add
```bash
--embedding_path /path/to/embeddings.pt
```
to load the result of your finetuning
Note: use --full_precision for 1x< gpus
### More commands
See [InvokeAI's docs](https://github.com/invoke-ai/InvokeAI)