app.py

from diffusers import StableDiffusionXLPipeline, AutoencoderKL
import torch
import random
import os
#from controlnet_aux import OpenposeDetector
#from diffusers.utils import load_image
import gradio as gr
import gc

model_id = int(os.getenv("Model"))

#stable-diffusion-xl-base-1.0  0 - base model
#Colossus_Project_XL           1 - better people
#Sevenof9_v3_sdxl              2 - nsfw
#JuggernautXL_version5         3 - better faces
#RealVisXL_V2.0                4 - realistic people
#AlbedoBaseXL_v11              5 - realistic people
#BetterThanWords_v20_sdxl      6 - nsfw
#AcornIsSpinning_acornXLV1     7 - nsfw

model_url_list = ["stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0.safetensors",
                 "Krebzonide/Colossus_Project_XL/blob/main/colossusProjectXLSFW_v202BakedVAE.safetensors",
                 "Krebzonide/Sevenof9_v3_sdxl/blob/main/nsfwSevenof9V3_nsfwSevenof9V3.safetensors",
                 "Krebzonide/JuggernautXL_version5/blob/main/juggernautXL_version5.safetensors",
                 "SG161222/RealVisXL_V2.0/blob/main/RealVisXL_V2.0.safetensors",
                 "Krebzonide/AlbedoBaseXL_v11/blob/main/albedobaseXL_v11.safetensors",
                 "Krebzonide/BetterThanWords_v20_sdxl/blob/main/betterThanWords_v20.safetensors",
                 "Krebzonide/AcornIsSpinning_acornXLV1/blob/main/acornIsSpinning_acornxlV1.safetensors"]

css = """
.btn-green {
  background-image: linear-gradient(to bottom right, #6dd178, #00a613) !important;
  border-color: #22c55e !important;
  color: #166534 !important;
}
.btn-green:hover {
  background-image: linear-gradient(to bottom right, #6dd178, #6dd178) !important;
}
"""

def generate(prompt, neg_prompt, samp_steps, guide_scale, batch_size, seed, height, width, progress=gr.Progress(track_tqdm=True)):
    if seed < 0:
        seed = random.randint(1,999999)
    images = pipe(
        prompt,
        negative_prompt=neg_prompt,
        num_inference_steps=samp_steps,
        guidance_scale=guide_scale,
        #cross_attention_kwargs={"scale": lora_scale},
        num_images_per_prompt=batch_size,
        height=height,
        width=width,
        generator=torch.manual_seed(seed),
    ).images
    return [(img, f"Image {i+1}") for i, img in enumerate(images)]
        
def set_base_model(base_model_id):
    vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
    global model_url_list
    model_url = "https://huggingface.co/" + model_url_list[base_model_id]
    pipe = StableDiffusionXLPipeline.from_single_file(
        model_url,
        torch_dtype = torch.float16,
        variant = "fp16",
        vae = vae,
        use_safetensors = True,
        use_auth_token="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"
    )
    pipe.to("cuda")
    return pipe

def update_pixel_ratio(height, width):
    return round(height*width/1048576,3)
    
with gr.Blocks(css=css) as demo:
    with gr.Column():
        prompt = gr.Textbox(label="Prompt")
        negative_prompt = gr.Textbox(label="Negative Prompt")
        submit_btn = gr.Button("Generate", elem_classes="btn-green")
        with gr.Row():
            samp_steps = gr.Slider(1, 50, value=20, step=1, label="Sampling steps")
            guide_scale = gr.Slider(1, 6, value=3, step=0.5, label="Guidance scale")
            batch_size = gr.Slider(1, 6, value=1, step=1, label="Batch size")
        with gr.Row():
            height = gr.Slider(label="Height", value=1024, minimum=512, maximum=2048, step=16)
            width = gr.Slider(label="Width", value=1024, minimum=512, maximum=2048, step=16)
        with gr.Row():
            pixels = gr.Number(label="Pixel Ratio", value=1, interactive=False)
            seed = gr.Number(label="Seed", value=-1, minimum=-1, precision=0)
        gallery = gr.Gallery(label="Generated images", height=800)
    submit_btn.click(generate, [prompt, negative_prompt, samp_steps, guide_scale, batch_size, seed, height, width], [gallery], queue=True)
    height.change(update_pixel_ratio, [height, width], [pixels], queue=False)
    width.change(update_pixel_ratio, [height, width], [pixels], queue=False)
    

pipe = set_base_model(model_id)
demo.launch(debug=True)