app.py

import gradio as gr
import numpy as np
import random
from diffusers import DiffusionPipeline
import torch
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from cog_sdxl.dataset_and_utils import TokenEmbeddingsHandler
from cog_sdxl.no_init import no_init_or_tensor
from diffusers.models.attention_processor import LoRAAttnProcessor2_0

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

pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
).to(device)

unet = pipe.unet

lora_path = hf_hub_download(repo_id="SvenN/sdxl-emoji", filename="lora.safetensors", repo_type="model")
embeddings_path = hf_hub_download(repo_id="SvenN/sdxl-emoji", filename="embeddings.pti", repo_type="model")

#### Loading LoRA keys into the UNet #### 
tensors = load_file(lora_path)
unet_lora_attn_procs = {}
name_rank_map = {}
for tk, tv in tensors.items():
    # up is N, d
    tensors[tk] = tv.half()
    if tk.endswith("up.weight"):
        proc_name = ".".join(tk.split(".")[:-3])
        r = tv.shape[1]
        name_rank_map[proc_name] = r

for name, attn_processor in unet.attn_processors.items():
    cross_attention_dim = (
        None
        if name.endswith("attn1.processor")
        else unet.config.cross_attention_dim
    )
    if name.startswith("mid_block"):
        hidden_size = unet.config.block_out_channels[-1]
    elif name.startswith("up_blocks"):
        block_id = int(name[len("up_blocks.")])
        hidden_size = list(reversed(unet.config.block_out_channels))[
            block_id
        ]
    elif name.startswith("down_blocks"):
        block_id = int(name[len("down_blocks.")])
        hidden_size = unet.config.block_out_channels[block_id]
    with no_init_or_tensor():
        module = LoRAAttnProcessor2_0(
            hidden_size=hidden_size,
            cross_attention_dim=cross_attention_dim,
            rank=name_rank_map[name],
        ).half()
    unet_lora_attn_procs[name] = module.to("cuda", non_blocking=True)

unet.set_attn_processor(unet_lora_attn_procs)
unet.load_state_dict(tensors, strict=False)
#### End loading LoRA keys into the UNet 

### Start loading Embeddings into the text encoder ### 
handler = TokenEmbeddingsHandler(
    [pipe.text_encoder, pipe.text_encoder_2], [pipe.tokenizer, pipe.tokenizer_2]
)
handler.load_embeddings(embeddings_path)
### End loading embeddings into the text encoder ### 

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):

    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
        
    generator = torch.Generator().manual_seed(seed)
    
    image = pipe(
        prompt = prompt, 
        negative_prompt = negative_prompt,
        guidance_scale = guidance_scale, 
        num_inference_steps = num_inference_steps, 
        width = width, 
        height = height,
        generator = generator,
        cross_attention_kwargs={"scale": 0.6},
    ).images[0] 
    
    return image

examples = [
    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
    "An astronaut riding a green horse",
    "A delicious ceviche cheesecake slice",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""
        # SDXL Emoji running on diffusers 0.25.0 with cog-sdxl custom classes
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            run_button = gr.Button("Run", scale=0)
        
        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):
            
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
                visible=False,
            )
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=10.0,
                    step=0.1,
                    value=7.5,
                )
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=50,
                )
        
        gr.Examples(
            examples = examples,
            inputs = [prompt]
        )

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

demo.queue().launch()