app.py

import gradio as gr
from diffusers import DiffusionPipeline
import torch
import os
from huggingface_hub import login

# 通过环境变量获取 Token
hf_token = os.getenv("hf")

# 使用 Hugging Face Token 登录
login(token=hf_token)

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "stabilityai/stable-diffusion-3.5-large"

if torch.cuda.is_available():
    torch_dtype = torch.bfloat16
else:
    torch_dtype = torch.float32

pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)

pipe.load_lora_weights("prithivMLmods/SD3.5-Large-Photorealistic-LoRA", weight_name="Photorealistic-SD3.5-Large-LoRA.safetensors")
pipe = pipe.to(device)
pipe.fuse_lora(lora_scale=1.0)

# 定义图像生成函数
# @spaces.GPU(duration=65) 这个装饰器指定 函数运行在 GPU 上，并限制运行时间 最多 65 秒。
def generate_image(prompt, negative_prompt=""):
    print(f"正向提示词: {prompt}")
    print(f"负向提示词: {negative_prompt}")
    # •	prompt: 正向提示词，用于控制生成的内容。
	# •	negative_prompt: 负向提示词，用于减少不想要的特性。
	# •	seed: 随机种子，相同的 seed + prompt 会生成相同的图片。
	# •	randomize_seed: 是否随机化种子，如果为 True，则随机生成 seed。
	# •	width / height: 图片宽高，默认为 1024×1024。
	# •	guidance_scale: 提示词引导强度，越高越严格按 prompt 生成，值太高可能会降低质量。
	# •	num_inference_steps: 推理步数，一般 40~50 比较好，步数越高质量越好，但生成速度更慢。
	# •	progress=gr.Progress(track_tqdm=True): Gradio 进度条，用于在前端显示进度。

    return pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,  # 传递负面提示词
        num_inference_steps=40,
        randomize_seed=True, 
        guidance_scale=4.5,
        width=1024,
        height=1024,
    ).images[0]
    # return pipe(prompt).images[0]

# 创建 Gradio 界面
# iface = gr.Interface(fn=generate_image, inputs="text", outputs="image")
iface = gr.Interface(
    fn=generate_image, 
    inputs=[
        gr.Textbox(label="正向提示词"),  
        gr.Textbox(label="负向提示词", placeholder="（可选）输入你不想要的元素")  
    ], 
    outputs="image"
)

# 启动界面
iface.launch()


# import gradio as gr

# gr.load("models/prithivMLmods/SD3.5-Large-Photorealistic-LoRA").launch()

# import gradio as gr

# gr.load(
#    "models/prithivMLmods/SD3.5-Large-Photorealistic-LoRA",
#    provider="hf-inference",
# ).launch()


# import gradio as gr
# import torch
# from diffusers import StableDiffusion3Pipeline
# import os
# from huggingface_hub import login

# # 获取Hugging Face Token
# hf_token = os.environ.get("HF_TOKEN")
# login(token=hf_token)

# # 加载模型并配置
# pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3.5-large", torch_dtype=torch.bfloat16)
# pipe.load_lora_weights("prithivMLmods/SD3.5-Large-Photorealistic-LoRA", weight_name="Photorealistic-SD3.5-Large-LoRA.safetensors")
# pipe.fuse_lora(lora_scale=1.0)

# # 如果有GPU，转移到GPU
# # pipe.to("cuda")

# # 定义图像生成函数，添加种子参数
# def generate_image(prompt, seed):
#     # 设置种子
#     generator = torch.manual_seed(seed)
    
#     # 使用模型生成图像
#     result = pipe(prompt=prompt,
#                   num_inference_steps=24, 
#                   guidance_scale=4.0,
#                   width=960, height=1280,
#                   generator=generator)
    
#     # 确保返回 PIL 图像
#     image = result.images[0]
#     print(type(image))
#     return image

# # 创建Gradio界面（使用 Interface）
# def gradio_interface():
#     with gr.Interface(fn=generate_image, 
#                       inputs=[gr.Textbox(label="Prompt", value="Man in the style of dark beige and brown, uhd image, youthful protagonists, nonrepresentational photography"),
#                               gr.Slider(minimum=0, maximum=100000, step=1, label="Seed", value=42)], 
#                       outputs=gr.Image(type="pil", label="Generated Image")) as demo:
#         demo.launch()

# # 启动Gradio应用
# gradio_interface()
    
# # 创建Gradio界面
# # with gr.Blocks() as demo:
# #     gr.Markdown("## Stable Diffusion Image Generation with Seed Control")

# #     # 输入框：提示文本
# #     prompt_input = gr.Textbox(label="Prompt", value="Man in the style of dark beige and brown, uhd image, youthful protagonists, nonrepresentational photography")

# #     # 滑块：种子
# #     seed_input = gr.Slider(minimum=0, maximum=100000, step=1, label="Seed", value=42)

# #     # 输出图像
# #     output_image = gr.Image(type="pil", label="Generated Image")

# #     # 按钮触发事件
# #     generate_btn = gr.Button("Generate Image")
# #     generate_btn.click(fn=generate_image, inputs=[prompt_input, seed_input], outputs=output_image)

# # # 启动Gradio应用
# # demo.launch()