t2i/StableCascade.py

# https://hf-mirror.com/stabilityai/stable-cascade
# https://hf-mirror.com/stabilityai/stable-cascade-prior

import torch
from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline, StableCascadeCombinedPipeline

cas = "stabilityai/stable-cascade"
cas_prior = "stabilityai/stable-cascade-prior"

def t2i_(prompt):

    prior = StableCascadePriorPipeline.from_pretrained(cas_prior, variant="bf16", torch_dtype=torch.bfloat16)
    decoder = StableCascadeDecoderPipeline.from_pretrained(cas, variant="bf16", torch_dtype=torch.float16)
    
    prior.to("cuda")
    decoder.to("cuda")
    # prior.enable_model_cpu_offload()
    # decoder.enable_model_cpu_offload()

    prior_output = prior(
        prompt=prompt,
        height=1024,
        width=1024,
        negative_prompt="",
        guidance_scale=4.0,
        num_images_per_prompt=1,
        num_inference_steps=20
    )
    
    image = decoder(
        image_embeddings=prior_output.image_embeddings.to(torch.float16),
        prompt=prompt,
        negative_prompt="",
        guidance_scale=0.0,
        output_type="pil",
        num_inference_steps=10
    ).images[0]

    return image

def t2i(prompt):
    pipe = StableCascadeCombinedPipeline.from_pretrained(cas, variant="bf16", torch_dtype=torch.bfloat16)
    pipe.to("cuda")
    
    image = pipe(
        prompt=prompt,
        negative_prompt="",
        num_inference_steps=10,
        prior_num_inference_steps=20,
        prior_guidance_scale=3.0,
        width=1024,
        height=1024,
    ).images[0]

    return image

if __name__ == "__main__":
    prompt = "a girl in beijing"
    image = t2i(prompt)
    # image = t2i_(prompt)
    image.save("stablecascade_output.png")