app.py

import spaces
import os
import gradio as gr
import torch
import numpy as np
import torch
from pytorch_lightning import seed_everything
from model_lib.utils import parse_args
from model_lib.modules import MoMA_main_modal
os.environ["CUDA_VISIBLE_DEVICES"]="0"

title = "MoMA"
description = "This demo is running on Zero_GPU with 4bit quantization. Please find our project page at https://moma-adapter.github.io"
device = torch.device('cuda')
seed_everything(0)

args = parse_args()

model = MoMA_main_modal(args).to(device, dtype=torch.float16)

@spaces.GPU(duration=120)
def inference(rgb, subject, prompt, strength, seed):
    global model
    seed = int(seed) if seed else 0
    seed = seed if not seed == 0 else np.random.randint(0,1000)
    generated_image = model.generate_images(rgb, subject, prompt, strength=strength, seed=seed)
    return generated_image

gr.Interface(
    inference,
    [gr.Image(type="pil", label="Input RGB"),
     gr.Textbox(lines=1, label="subject"),
     gr.Textbox(lines=1, label="Prompt"),
     gr.Slider(minimum=0.2, maximum=1.2, step=0.1,label="Strength. Recommend: 1.0 for context editing; 0.4 for texture editing",value=1.0),
     gr.Textbox(lines=1, label="Seed. Use 0 for a random seed")],
    gr.Image(type="pil", label="Output"),
    title=title,
    description=description,
    examples=[["example_images/newImages/3.jpg",'car','A car in autumn with falling leaves.',1.0,"6"],["example_images/newImages/3.jpg",'car','A wooden sculpture of a car on a table.',0.4,"4"],["example_images/newImages/2.jpg",'car','A car on a city road with green trees and buildings.',1.0,"4"],["example_images/newImages/03.jpg",'cat','A cat at the Grand Canyon.',1.0,"2"],["example_images/newImages/02.jpg",'dog','A dog in a spring garden with flowers.',1.0,"6"],["example_images/newImages/1.jpeg",'bird','A bird in spring with flowers.',1.0,"1"],["example_images/newImages/17.jpg",'robot','A robot in autumn mountain and lake.',1,"5"]],
    allow_flagging='never'
).launch(debug=False)