stable_diffusion.openvino

Implementation of Text-To-Image generation using Stable Diffusion on Intel CPU.

News

When we started this project, it was just a tiny proof of concept that you can work with state-of-the-art image generators even without access to expensive hardware. But, due we get a lot of feedback from you, we decided to make this project something more than a tiny script. Currently, we work on the new version of our project, so we can respond to your issues and pool requests with delay.

Requirements

Linux, Windows, MacOS
Python 3.8.+
CPU compatible with OpenVINO.

Install requirements

pip install -r requirements.txt

Generate image from text description

usage: demo.py [-h] [--model MODEL] [--seed SEED] [--beta-start BETA_START] [--beta-end BETA_END] [--beta-schedule BETA_SCHEDULE] [--num-inference-steps NUM_INFERENCE_STEPS]
               [--guidance-scale GUIDANCE_SCALE] [--eta ETA] [--tokenizer TOKENIZER] [--prompt PROMPT] [--params-from PARAMS_FROM] [--init-image INIT_IMAGE] 
               [--strength STRENGTH] [--mask MASK] [--output OUTPUT]

optional arguments:
  -h, --help            show this help message and exit
  --model MODEL         model name
  --seed SEED           random seed for generating consistent images per prompt
  --beta-start BETA_START
                        LMSDiscreteScheduler::beta_start
  --beta-end BETA_END   LMSDiscreteScheduler::beta_end
  --beta-schedule BETA_SCHEDULE
                        LMSDiscreteScheduler::beta_schedule
  --num-inference-steps NUM_INFERENCE_STEPS
                        num inference steps
  --guidance-scale GUIDANCE_SCALE
                        guidance scale
  --eta ETA             eta
  --tokenizer TOKENIZER
                        tokenizer
  --prompt PROMPT       prompt
  --params-from PARAMS_FROM
                        Extract parameters from a previously generated image.
  --init-image INIT_IMAGE
                        path to initial image
  --strength STRENGTH   how strong the initial image should be noised [0.0, 1.0]
  --mask MASK           mask of the region to inpaint on the initial image
  --output OUTPUT       output image name

Examples

Example Text-To-Image

python demo.py --prompt "Street-art painting of Emilia Clarke in style of Banksy, photorealism"

Repeat a previous image generation with identical seed and parameters but more steps

python demo.py --params-from output.png --output new-output.png --num-inference-steps 64

Example Image-To-Image

python demo.py --prompt "Photo of Emilia Clarke with a bright red hair" --init-image ./data/input.png --strength 0.5

Example Inpainting

python demo.py --prompt "Photo of Emilia Clarke with a bright red hair" --init-image ./data/input.png --mask ./data/mask.png --strength 0.5

Example web demo

Example video on YouTube

streamlit run demo_web.py

Performance

CPU	Time per iter	Total time
AMD Ryzen Threadripper 1900X	5.34 s/it	2.58 min
Intel(R) Core(TM) i7-4790K @ 4.00GHz	10.1 s/it	5.39 min
Intel(R) Core(TM) i5-8279U	7.4 s/it	3.59 min
Intel(R) Core(TM) i7-1165G7 @ 2.80GHz	7.4 s/it	3.59 min
Intel(R) Core(TM) i7-11800H @ 2.30GHz (16 threads)	2.9 s/it	1.54 min
Intel(R) Xeon(R) Gold 6154 CPU @ 3.00GHz	1 s/it	33 s

Acknowledgements

Original implementation of Stable Diffusion: https://github.com/CompVis/stable-diffusion
diffusers library: https://github.com/huggingface/diffusers

Disclaimer

The authors are not responsible for the content generated using this project. Please, don't use this project to produce illegal, harmful, offensive etc. content.

senthilnayagam / stable_diffusion.openvino-imagemetadata