app.py

import spaces
import os
import torch
import random
from huggingface_hub import snapshot_download
from kolors.pipelines.pipeline_stable_diffusion_xl_chatglm_256 import StableDiffusionXLPipeline
from kolors.models.modeling_chatglm import ChatGLMModel
from kolors.models.tokenization_chatglm import ChatGLMTokenizer
from diffusers import UNet2DConditionModel, AutoencoderKL
from diffusers import EulerDiscreteScheduler
import gradio as gr

# Download the model files
ckpt_dir = snapshot_download(repo_id="Kwai-Kolors/Kolors")

# Load the models
text_encoder = ChatGLMModel.from_pretrained(
    os.path.join(ckpt_dir, 'text_encoder'),
    torch_dtype=torch.float16).half()
tokenizer = ChatGLMTokenizer.from_pretrained(os.path.join(ckpt_dir, 'text_encoder'))
vae = AutoencoderKL.from_pretrained(os.path.join(ckpt_dir, "vae"), revision=None).half()
scheduler = EulerDiscreteScheduler.from_pretrained(os.path.join(ckpt_dir, "scheduler"))
unet = UNet2DConditionModel.from_pretrained(os.path.join(ckpt_dir, "unet"), revision=None).half()

pipe = StableDiffusionXLPipeline(
        vae=vae,
        text_encoder=text_encoder,
        tokenizer=tokenizer,
        unet=unet,
        scheduler=scheduler,
        force_zeros_for_empty_prompt=False)
pipe = pipe.to("cuda")



import gradio as gr
import numpy as np
import random
import torch
from diffusers import AutoPipelineForText2Image

import spaces

device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float16


repo = "SG161222/RealVisXL_V4.0"

pipeline_real = AutoPipelineForText2Image.from_pretrained(repo, torch_dtype=torch.float16).to('cuda')  


def adjust_to_nearest_multiple(value, divisor=8):
    """
    Adjusts the input value to the nearest multiple of the divisor.
    
    Args:
    value (int): The value to adjust.
    divisor (int): The divisor to which the value should be divisible. Default is 8.
    Returns:
    int: The nearest multiple of the divisor.
    """
    if value % divisor == 0:
        return value
    else:
        # Round to the nearest multiple of divisor
        return round(value / divisor) * divisor

def adjust_dimensions(height, width):
    """
    Adjusts the height and width to be divisible by 8.
    
    Args:
    height (int): The height to adjust.
    width (int): The width to adjust.
    Returns:
    tuple: Adjusted height and width.
    """
    new_height = adjust_to_nearest_multiple(height)
    new_width = adjust_to_nearest_multiple(width)
    
    return new_height, new_width


MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 4100


@spaces.GPU(duration=100)
def generate_image(prompt, negative_prompt, height, width, num_inference_steps, guidance_scale, num_images_per_prompt, use_random_seed, seed, progress=gr.Progress(track_tqdm=True)):
    if use_random_seed:
        seed = random.randint(0, 2**32 - 1)
    else:
        seed = int(seed)  # Ensure seed is an integer

    width = min(width, MAX_IMAGE_SIZE // 2)
    height = min(height, MAX_IMAGE_SIZE // 2)
    height, width = adjust_dimensions(height, width)
    
    if negative_prompt=="1":
        image = pipe(
            prompt=prompt,
            negative_prompt="",
            height=height,
            width=width,
            num_inference_steps=num_inference_steps,
            guidance_scale=guidance_scale,
            num_images_per_prompt=num_images_per_prompt,
            generator=torch.Generator(pipe.device).manual_seed(seed)
        ).images
        return image, seed
    
    generator = torch.Generator().manual_seed(seed)
    image = pipeline_real(prompt = prompt, 
        negative_prompt = "",
        guidance_scale = guidance_scale, 
        num_inference_steps = num_inference_steps, 
        width = width, 
        height = height,
        generator = generator
                    ).images

    return image, seed

description = """
<p align="center">Effective Training of Diffusion Model for Photorealistic Text-to-Image Synthesis</p>
<p><center>
<a href="https://kolors.kuaishou.com/" target="_blank">[Official Website]</a>
<a href="https://github.com/Kwai-Kolors/Kolors/blob/master/imgs/Kolors_paper.pdf" target="_blank">[Tech Report]</a>
<a href="https://huggingface.co/Kwai-Kolors/Kolors" target="_blank">[Model Page]</a>
<a href="https://github.com/Kwai-Kolors/Kolors" target="_blank">[Github]</a>
</center></p>
"""

# Gradio interface
iface = gr.Interface(
    fn=generate_image,
    inputs=[
        gr.Textbox(label="Prompt"),
        gr.Textbox(label="Negative Prompt")
    ],
    additional_inputs=[
        gr.Slider(512, 2048, 1024, step=64, label="Height"),
        gr.Slider(512, 2048, 1024, step=64, label="Width"),
        gr.Slider(20, 50, 20, step=1, label="Number of Inference Steps"),
        gr.Slider(1, 20, 5, step=0.5, label="Guidance Scale"),
        gr.Slider(1, 4, 1, step=1, label="Number of images per prompt"),
        gr.Checkbox(label="Use Random Seed", value=True),
        gr.Number(label="Seed", value=0, precision=0)
    ],
    additional_inputs_accordion=gr.Accordion(label="Advanced settings", open=False),
    outputs=[
        gr.Gallery(label="Result", elem_id="gallery", show_label=False),
        gr.Number(label="Seed Used")
    ],
    title="Kolors",
    description=description,
    theme='bethecloud/storj_theme',
)

iface.launch(debug=True)