Download the files all from huggingface and place them into the "local" folder. There are two models available:
- https://huggingface.co/bes-dev/stable-diffusion-v1-4-openvino/tree/main
- https://huggingface.co/ShadowPower/waifu-diffusion.openvino/tree/main
Extract the zip file and delete the zip file. You should have a "venv" folder alongside run.bat and other files.
Double click run.bat.
Write a prompt. Enter this in the terminal that pops up as an example:
python demo.py --prompt "frog on a log" --num-inference-steps 8
Select sampler(scheduler) and more:
python demo.py --prompt "a photo of a dog" --unprompt "trees, people, low quality" --scheduler DDIM --num-inference-steps 8 --seed 5
After running for the first time, huggingface caches files in username.cache\huggingface\transformers.
To fully use offline change the 0 to 1 in: venv/lib/python3.9/site-packages/transformers/file_utils.py line 328:
_is_offline_mode = True if os.environ.get("TRANSFORMERS_OFFLINE", "0").upper() in ENV_VARS_TRUE_VALUES else False
This gets rid of the error when you are not connected to wifi.
This repository does two things: It makes the codebase and models portable and offline for use with usb, ssds. It allows DDIM sampler and negative prompts, as taken from Drake53's repository. DDIM sampler often converges at lower steps (8), making image generation alot quicker.
This needs 16gb of ram to run smoothly. If you are using 8 gb, you will end up using ROM storage of your hard drive or solid state drive. It is highly discouraged to use (spinning disk)hard drive, tests show 126 seconds per iteration step. For 8 gb ram users: see this page
This repository code is from https://github.com/Drake53/stable_diffusion.openvino and is a fork of https://github.com/bes-dev/stable_diffusion.openvino