app.py

# Ref: https://huggingface.co/spaces/multimodalart/cosxl
import gradio as gr
from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
import spaces 
import torch 
import os

from compel import Compel, ReturnedEmbeddingsType

from huggingface_hub import hf_hub_download
from safetensors.torch import load_file

model_id = "aipicasso/emi-2"
token=os.environ["TOKEN"]

scheduler = EulerAncestralDiscreteScheduler.from_pretrained(model_id,subfolder="scheduler",token=token)
pipe_normal = StableDiffusionXLPipeline.from_pretrained(model_id, scheduler=scheduler, torch_dtype=torch.bfloat16,token=token)

negative_ti_file = hf_hub_download(repo_id="Aikimi/unaestheticXL_Negative_TI", filename="unaestheticXLv31.safetensors")
state_dict = load_file(negative_ti_file)
pipe_normal.load_textual_inversion(state_dict["clip_g"], token="unaestheticXLv31", text_encoder=pipe_normal.text_encoder_2, tokenizer=pipe_normal.tokenizer_2)
pipe_normal.load_textual_inversion(state_dict["clip_l"], token="unaestheticXLv31", text_encoder=pipe_normal.text_encoder, tokenizer=pipe_normal.tokenizer)

state_dict = load_file("unaestheticXL_Alb2.safetensors")
pipe_normal.load_textual_inversion(state_dict["clip_g"], token="unaestheticXL_Alb2", text_encoder=pipe_normal.text_encoder_2, tokenizer=pipe_normal.tokenizer_2)
pipe_normal.load_textual_inversion(state_dict["clip_l"], token="unaestheticXL_Alb2", text_encoder=pipe_normal.text_encoder, tokenizer=pipe_normal.tokenizer)

pipe_normal.load_lora_weights("fix_hands.pt")
pipe_normal.fuse_lora(lora_scale=1.0)

pipe_normal.to("cuda")

pipe_normal.enable_freeu(s1=1.2, s2=0.7, b1=1.1, b2=1.3)

compel = Compel(tokenizer=[pipe_normal.tokenizer, pipe_normal.tokenizer_2] , 
                text_encoder=[pipe_normal.text_encoder, pipe_normal.text_encoder_2], 
                returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED, 
                requires_pooled=[False, True])

@spaces.GPU
def run_normal(prompt, negative_prompt="", guidance_scale=7.5, progress=gr.Progress(track_tqdm=True)):
    # ユーザーの著作権侵害を防ぐフィルター
    words=["pokemon", "pikachu", "picachu", "mario", "sonic", "genshin"]
    for word in words:
        prompt=prompt.replace(word,"")
    if(prompt==""):
        conditioning, pooled = compel("1girl, (upper body)++, brown bob short hair, brown eyes, looking at viewer, cherry blossom")
    else:
        conditioning, pooled = compel(prompt)
    negative_conditioning, negatice_pooled = compel("(unaestheticXLv31)++++, (unaestheticXL_Alb2)++++, bad hands, bad anatomy, low quality, 3d, photo, realism, text, sign, "+negative_prompt)
    result = pipe_normal(
        prompt_embeds=conditioning,
        pooled_prompt_embeds=pooled, 
        negative_prompt_embeds=negative_conditioning, 
        negative_pooled_prompt_embeds=negatice_pooled,
        num_inference_steps = 25,
        guidance_scale = guidance_scale,
        width = 768,
        height = 1344)
    
    return result.images[0]

css = '''
.gradio-container{
max-width: 768px !important;
margin: 0 auto;
}
'''

normal_examples = [
    "1girl, (upper body)++, brown bob short hair, brown eyes, looking at viewer, cherry blossom",
    "1girl, (full body)++, brown bob short hair, brown eyes, school uniform, cherry blossom",
    "no humans, manga, black and white, monochrome, Mt. fuji, 4k, highly detailed",
    "no humans, manga, black and white, monochrome, Shibuya street, 4k, highly detailed",
    "1boy, (upper body)++, silver very short hair, red eyes, looking at viewer, white background",
    "1boy, (full body)++, silver very short hair, red eyes, looking at viewer, white background",
]

with gr.Blocks(css=css) as demo:
    gr.Markdown('''# Emi 2
    Official demo for [Emi 2](https://huggingface.co/aipicasso/emi-2). Click the generate button!<br>
    本モデルの生成物は各種法令に従って取り扱って下さい。
    ''')
    with gr.Group():
        with gr.Row():
          prompt_normal = gr.Textbox(show_label=False, scale=4, placeholder="Your prompt, e.g.: 1girl, (upper body)++, brown bob short hair, brown eyes, looking at viewer, cherry blossom")
          button_normal = gr.Button("Generate", min_width=120)
        output_normal = gr.Image(label="Your result image", interactive=False)
        with gr.Accordion("Advanced Settings", open=False):
          negative_prompt_normal = gr.Textbox(label="Negative Prompt")
          guidance_scale_normal = gr.Number(label="Guidance Scale", value=7.5)
    gr.Examples(examples=normal_examples, fn=run_normal, inputs=[prompt_normal], outputs=[output_normal], cache_examples=True) 
    
    gr.on(
        triggers=[
            button_normal.click,
            prompt_normal.submit
        ],
        fn=run_normal,
        inputs=[prompt_normal, negative_prompt_normal, guidance_scale_normal],
        outputs=[output_normal],
    )
if __name__ == "__main__":
    demo.launch(share=True)