Guidance scale

Guidance scale or Classifier-Free Guidance (CFG) is a parameter that controls the adherence of the generated image to the text prompt. A higher guidance scale will force the image to be more aligned with the prompt, while a lower guidance scale will give more space for Stable Diffusion to decide what to put into the image.

Here is a sample of applying a different guidance scale while keeping other parameters the same:

import torch
generator = torch.Generator("cuda:0").manual_seed(123)
prompt = """high resolution, a photograph of an astronaut riding a horse on mars"""
image_3_gs = text2img_pipe(
    prompt = prompt,
    num_inference_steps = 30,
    guidance_scale = 3,
    generator = generator
).images[0]
image_7_gs = text2img_pipe(
    prompt = prompt,
    num_inference_steps = 30,
    ...