app.py

import gradio as gr
import torch, random, time
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline
from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
translations = {
	'en': {
		'model_name': 'Model Path',
		'loading': 'Loading',
		'input': 'Input Image',
		'prompt': 'Prompt',
		'negative_prompt': 'Negative Prompt',
		'generate': 'Generate',
		'strength': 'Strength',
		'scale': 'Guidance Scale',
		'steps': 'Number of Inference Steps',
		'width': 'Width',
		'height': 'Height',
		'seed': 'Seed',
	},
	'zh': {
		'model_name': '模型路径',
		'loading': '载入',
		'input': '输入图像',
		'prompt': '提示',
		'negative_prompt': '负面提示',
		'generate': '生成',
		'strength': '强度',
		'scale': '引导系数',
		'steps': '迭代步数',
		'width': '宽度',
		'height': '高度',
		'seed': '机数种子',
	}
}
language='zh'
def generate_new_seed():
	return random.randint(1, 2147483647)
def update_language(new_language):
	return [
		gr.Textbox.update(label=translations[new_language]['model_name']),
		gr.Button.update(value=translations[new_language]['loading']),
		gr.Image.update(label=translations[new_language]['input']),
		gr.Textbox.update(label=translations[new_language]['prompt']),
		gr.Textbox.update(label=translations[new_language]['negative_prompt']),
		gr.Button.update(value=translations[new_language]['generate']),
		gr.Slider.update(label=translations[new_language]['strength']),
		gr.Slider.update(label=translations[new_language]['scale']),
		gr.Slider.update(label=translations[new_language]['steps']),
		gr.Slider.update(label=translations[new_language]['width']),
		gr.Slider.update(label=translations[new_language]['height']),
		gr.Number.update(label=translations[new_language]['seed'])
	]

text2img = None
img2img = None
def Generate(image_input, prompt, negative_prompt, strength, scale, steps, width, height, seed):
	global text2img, img2img
	if seed == -1:
		seed = generate_new_seed()
	generator = torch.Generator().manual_seed(int(seed))
	start_time = time.time()
	if image_input is None:
		image = text2img(prompt=prompt, negative_prompt=negative_prompt, guidance_scale=scale, num_inference_steps=steps, width=width, height=height, num_images_per_prompt=1, generator=generator).images[0]
	else:
		image = img2img(image=image_input, strength=0.75, prompt=prompt, negative_prompt=negative_prompt, guidance_scale=scale, num_inference_steps=steps, width=width, height=height, num_images_per_prompt=1, generator=generator).images[0]
	minutes, seconds = divmod(round(time.time() - start_time), 60)
	return image, f"seed:{seed}\ntime:{minutes:02d}:{seconds:02d}"
def Loading(model_name, is_xl, is_cuda):
	global text2img, img2img
	if is_xl == False:
		is_xl ='xl' in model_name.lower()
	device = "cuda" if is_cuda else "cpu"
	pipeline_class = StableDiffusionXLPipeline if is_xl else StableDiffusionPipeline
	if is_cuda:
		text2img = pipeline_class.from_pretrained(model_name, torch_dtype=torch.float16, variant="fp16", use_safetensors=True).to(device)
	else:
		text2img = pipeline_class.from_pretrained(model_name, use_safetensors=True).to(device)
	if is_cuda:
		text2img.enable_xformers_memory_efficient_attention()
		text2img.vae.enable_xformers_memory_efficient_attention()
	text2img.safety_checker = None
	img2img = (StableDiffusionXLImg2ImgPipeline if is_xl else StableDiffusionImg2ImgPipeline)(**text2img.components)
	return model_name
with gr.Blocks() as demo:
	with gr.Row():
		model_name = gr.Textbox(value="nota-ai/bk-sdm-tiny-2m", label=translations[language]['model_name'])
		with gr.Column():
			is_xl = gr.Checkbox(label="SDXL")
			is_cuda = gr.Checkbox(label="cuda", value=torch.cuda.is_available())
		loading = gr.Button(translations[language]['loading'], variant='primary')
		set_language = gr.Dropdown(list(translations.keys()), label="Language", value=language)
	with gr.Row():
		with gr.Column(scale=2):
			with gr.Column():
				image_input = gr.Image(label=translations[language]['input'])
				strength = gr.Slider(minimum=0, maximum=1, value=0.8, step=0.01, label=translations[language]['strength'])
			prompt = gr.Textbox("space warrior, beautiful, female, ultrarealistic, soft lighting, 8k", label=translations[language]['prompt'], lines=3)
			negative_prompt = gr.Textbox("deformed, distorted, disfigured, poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, mutated hands and fingers, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation,lowres,jpeg artifacts,username,logo,signature,watermark,monochrome,greyscale", label=translations[language]['negative_prompt'], lines=3)
			with gr.Row():
				width = gr.Slider(minimum=512, maximum=2048, value=512, step=8, label=translations[language]['width'])
				height = gr.Slider(minimum=512, maximum=2048, value=512, step=8, label=translations[language]['height'])
			with gr.Row():
				scale = gr.Slider(minimum=1, maximum=15, value=7.5, step=0.5, label=translations[language]['scale'])
				steps = gr.Slider(minimum=1, maximum=100, value=50, step=1, label=translations[language]['steps'])
			with gr.Row():
				seed = gr.Slider(minimum=-1, step=1, maximum=2147483647, value=-1, label=translations[language]['seed'])
				set_seed = gr.Button("🎲")
		with gr.Column(scale=3):
			image_output = gr.Image()
			text_output = gr.Textbox()
			generate = gr.Button(translations[language]['generate'], variant='primary')
	set_seed.click(generate_new_seed, None, seed)
	generate.click(Generate, [image_input, prompt, negative_prompt, strength, scale, steps, width, height, seed], [image_output, text_output])
	loading.click(Loading, [model_name, is_xl, is_cuda], model_name)
	set_language.change(update_language, set_language, [model_name, loading, image_input, prompt, negative_prompt, generate, strength, scale, steps, width, height, seed])
demo.queue().launch()