Document parameter defaults

[xeruf]

Especially these two seem interesting

  --num-inference-steps NUM_INFERENCE_STEPS
                        num inference steps
  --guidance-scale GUIDANCE_SCALE
                        guidance scale

By limiting the inference steps, you seem to get an intermediate result faster. Is there any other way to get a quick prototype, maybe in lower resolution or something?

[rncar]

The resolution is always 512x512, is the output of the neural net. The only way to go faster is limiting num-inference-steps.

You can see the parameter defaults in demo.py:

--model default="bes-dev/stable-diffusion-v1-4-openvino"
--seed default=None # none means random
--beta-start default=0.00085
--beta-end default=0.012
--beta-schedule default="scaled_linear"
--num-inference-steps default=32
--guidance-scale default=7.5
--eta default=0.0
--tokenizer default="openai/clip-vit-large-patch14"
--prompt default="Street-art painting of Emilia Clarke in style of Banksy, photorealism"
--init-image default=None
--strength default=0.5
--mask" default=None
--output default="output.png"