app.py

import gradio as gr
import torch
from PIL import Image
from diffusers import AutoPipelineForText2Image, DDIMScheduler
from transformers import CLIPVisionModelWithProjection
import numpy as np
import spaces


image_encoder = CLIPVisionModelWithProjection.from_pretrained(
    "h94/IP-Adapter",
    subfolder="models/image_encoder",
    torch_dtype=torch.float16,
)

pipeline = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    image_encoder=image_encoder,
)
pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)

pipeline.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name=["ip-adapter-plus_sdxl_vit-h.safetensors", "ip-adapter-plus-face_sdxl_vit-h.safetensors"])
pipeline.set_ip_adapter_scale([0.7, 0.5])

pipeline.enable_model_cpu_offload()

@spaces.GPU
def transform_image(face_image):
    generator = torch.Generator(device="cpu").manual_seed(0)

    # Check if the input is already a PIL Image
    if isinstance(face_image, Image.Image):
        processed_face_image = face_image
    # If the input is a NumPy array, convert it to a PIL Image
    elif isinstance(face_image, np.ndarray):
        processed_face_image = Image.fromarray(face_image)
    else:
        raise ValueError("Unsupported image format")

    # Load the style image from the local path
    style_image_path = "examples/soyjak2.jpg"
    style_image = Image.open(style_image_path)

    # Perform the transformation
    image = pipeline(
        prompt="soyjak",
        ip_adapter_image=[style_image, processed_face_image],
        negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
        num_inference_steps=30,
        generator=generator,
    ).images[0]

    return image

# Gradio interface setup
demo = gr.Interface(
    fn=transform_image,
    inputs=gr.Image(label="Upload your face image"),
    outputs=gr.Image(label="Your Soyjak"),
    title="InstaSoyjak - turn anyone into a Soyjak",
    description="All you need to do is upload an image. Please use responsibly. Please follow me on Twitter if you like this space: https://twitter.com/angrypenguinPNG. Idea from Yacine, please give him a follow: https://twitter.com/yacineMTB.",
)

demo.queue(max_size=20)  # Configures the queue with a maximum size of 20
demo.launch()