import random
import os
import gradio as gr
import numpy as np
import spaces
import torch
from diffusers import AutoPipelineForText2Image, AutoencoderKL #,EulerDiscreteScheduler
from compel import Compel, ReturnedEmbeddingsType

if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>你现在运行在CPU上 但是此项目只支持GPU.</p>"

MAX_SEED = np.iinfo(np.int32).max
CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "1") == "1"
MAX_IMAGE_SIZE = 4096

if torch.cuda.is_available():
    vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
    pipe = AutoPipelineForText2Image.from_pretrained(
        "John6666/noobai-xl-nai-xl-epsilonpred075version-sdxl",
        vae=vae,
        torch_dtype=torch.float16,
        use_safetensors=True,
        add_watermarker=False
    )
    #pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
    pipe.to("cuda")

def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed

@spaces.GPU
def infer(
    prompt: str,
    negative_prompt: str = "lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]",
    use_negative_prompt: bool = True,
    seed: int = 7,
    width: int = 1024,
    height: int = 1536,
    guidance_scale: float = 3,
    num_inference_steps: int = 30,
    randomize_seed: bool = True,
    use_resolution_binning: bool = True,
    progress=gr.Progress(track_tqdm=True),
):
    seed = int(randomize_seed_fn(seed, randomize_seed))
    generator = torch.Generator().manual_seed(seed)
    compel = Compel(tokenizer=[pipe.tokenizer, pipe.tokenizer_2] , text_encoder=[pipe.text_encoder, pipe.text_encoder_2], returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED, requires_pooled=[False, True])
    conditioning, pooled = compel(prompt)
    
    image = pipe(
        #prompt=prompt,
        prompt_embeds=conditioning,
        pooled_prompt_embeds=pooled,
        negative_prompt=negative_prompt,
        width=width,
        height=height,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=generator,
        use_resolution_binning=use_resolution_binning,
    ).images[0]
    return image, seed

examples = [
    "nahida (genshin impact)",
    "klee (genshin impact)",
]

css = '''
.gradio-container{max-width: 560px !important}
h1{text-align:center}
footer {
    visibility: hidden
}
'''
    
with gr.Blocks(css=css) as demo:
    gr.Markdown("""# 梦羽的模型生成器
        ### 快速生成NoobAIXL v0.75的模型图片""")
    with gr.Group():
        with gr.Row():
            prompt = gr.Text(
                label="关键词",
                show_label=False,
                max_lines=1,
                placeholder="输入你要的图片关键词",
                container=False,
            )
            run_button = gr.Button("生成", scale=0, variant="primary")
        result = gr.Image(label="Result", show_label=False)
    with gr.Accordion("高级选项", open=False):
        with gr.Row():
            use_negative_prompt = gr.Checkbox(label="使用反向词条", value=True)
            negative_prompt = gr.Text(
                label="反向词条",
                max_lines=5,
                lines=4,
                placeholder="输入你要排除的图片关键词",
                value="lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]",
                visible=True,
            )
        seed = gr.Slider(
            label="种子",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
        )
        randomize_seed = gr.Checkbox(label="随机种子", value=True)
        with gr.Row(visible=True):
            width = gr.Slider(
                label="宽度",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1024,
            )
            height = gr.Slider(
                label="高度",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1536,
            )
        with gr.Row():
            guidance_scale = gr.Slider(
                label="Guidance Scale",
                minimum=0.1,
                maximum=10,
                step=0.1,
                value=7.0,
            )
            num_inference_steps = gr.Slider(
                label="生成步数",
                minimum=1,
                maximum=50,
                step=1,
                value=28,
            )

    gr.Examples(
        examples=examples,
        inputs=prompt,
        outputs=[result, seed],
        fn=infer,
        cache_examples=CACHE_EXAMPLES,
    )
    
    use_negative_prompt.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_negative_prompt,
        outputs=negative_prompt,
    )

    gr.on(
        triggers=[prompt.submit,run_button.click],
        fn=infer,
        inputs=[
            prompt,
            negative_prompt,
            use_negative_prompt,
            seed,
            width,
            height,
            guidance_scale,
            num_inference_steps,
            randomize_seed,
        ],
        outputs=[result, seed],
    )

if __name__ == "__main__":
    demo.launch()