app.py

import gradio as gr
import numpy as np
import random
import gc
import json
import torch
import spaces

from huggingface_hub import hf_hub_download
from diffusers import (
    AutoencoderKL,
    SD3Transformer2DModel,
    StableDiffusion3Pipeline,
    FlowMatchEulerDiscreteScheduler
)
from diffusers.loaders.single_file_utils import (
    convert_sd3_transformer_checkpoint_to_diffusers,
)
from transformers import (
    CLIPTextModelWithProjection,
    CLIPTokenizer,
    T5EncoderModel,
    T5Tokenizer
)
from accelerate import init_empty_weights
from accelerate.utils import set_module_tensor_to_device
from safetensors import safe_open

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "stabilityai/stable-diffusion-3.5-large"
finetune_repo_id = "DoctorDiffusion/Absynth-2.0"
finetune_filename = "Absynth_SD3.5L_2.0.safetensors"

if torch.cuda.is_available():
    torch_dtype = torch.bfloat16
else:
    torch_dtype = torch.float32

# Initialize transformer
config_file = hf_hub_download(repo_id=model_repo_id, filename="transformer/config.json")
with open(config_file, "r") as fp:
    config = json.load(fp)
with init_empty_weights():
    transformer = SD3Transformer2DModel.from_config(config)

# Get transformer state dict and load
model_file = hf_hub_download(repo_id=finetune_repo_id, filename=finetune_filename)
state_dict = {}
with safe_open(model_file, framework="pt") as f:
    for key in f.keys():
        state_dict[key] = f.get_tensor(key)

state_dict = convert_sd3_transformer_checkpoint_to_diffusers(state_dict)
for key, value in state_dict.items():
    set_module_tensor_to_device(
        transformer,
        key,
        device,
        value=value,
        dtype=torch_dtype
    )

# Try to keep memory usage down
del state_dict
gc.collect()

# Initialize models from base SD3.5
vae = AutoencoderKL.from_pretrained(model_repo_id, subfolder="vae")
text_encoder = CLIPTextModelWithProjection.from_pretrained(model_repo_id, subfolder="text_encoder")
text_encoder_2 = CLIPTextModelWithProjection.from_pretrained(model_repo_id, subfolder="text_encoder_2")
text_encoder_3 = T5EncoderModel.from_pretrained(model_repo_id, subfolder="text_encoder_3")
tokenizer = CLIPTokenizer.from_pretrained(model_repo_id, subfolder="tokenizer")
tokenizer_2 = CLIPTokenizer.from_pretrained(model_repo_id, subfolder="tokenizer_2")
tokenizer_3 = T5Tokenizer.from_pretrained(model_repo_id, subfolder="tokenizer_3")
scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(model_repo_id, subfolder="scheduler")

# Create pipeline from our models
pipe = StableDiffusion3Pipeline(
    vae=vae,
    scheduler=scheduler,
    text_encoder=text_encoder,
    text_encoder_2=text_encoder_2,
    text_encoder_3=text_encoder_3,
    tokenizer=tokenizer,
    tokenizer_2=tokenizer_2,
    tokenizer_3=tokenizer_3,
    transformer=transformer
)
pipe = pipe.to(device, dtype=torch_dtype)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1536

@spaces.GPU(duration=65)
def infer(
    prompt,
    negative_prompt="",
    seed=42,
    randomize_seed=False,
    width=1024,
    height=1024,
    guidance_scale=4.5,
    num_inference_steps=40,
    progress=gr.Progress(track_tqdm=True),
):
    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,
    ).images[0]

    return image, seed


examples = [
    "An astrounaut encounters an alien on the moon, photograph",
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(" # [Absynth 2.0](https://huggingface.co/DoctorDiffusion/Absynth-2.0) by [DoctorDiffusion](https://civitai.com/user/doctor_diffusion)")
        gr.Markdown("Finetuned from [Stable Diffusion 3.5 Large (8B)](https://huggingface.co/stabilityai/stable-diffusion-3.5-large) by [Stability AI](https://stability.ai/news/introducing-stable-diffusion-3-5).")
        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, variant="primary")

        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",
            )

            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=512,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=768, 
                )

                height = gr.Slider(
                    label="Height",
                    minimum=512,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1344,
                )

            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=7.5,
                    step=0.1,
                    value=4.5,
                )

                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=40, 
                )

        gr.Examples(examples=examples, inputs=[prompt], outputs=[result, seed], fn=infer, cache_examples=True, cache_mode="lazy")

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

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