app.py

import gradio as gr
import numpy as np
import random
# import spaces
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL
from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images
from gradio_imageslider import ImageSlider
from PIL import Image, ImageDraw, ImageFont

dtype = torch.bfloat16
#model_id = "black-forest-labs/FLUX.1-dev"
model_id = "camenduru/FLUX.1-dev-diffusers"
device = "cuda" if torch.cuda.is_available() else "cpu"

#taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
good_vae = AutoencoderKL.from_pretrained(model_id, subfolder="vae", torch_dtype=dtype).to(device)
#pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=dtype, vae=taef1).to(device)
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=dtype, vae=good_vae).to(device)
torch.cuda.empty_cache()

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)

def get_cmp_image(im1: Image.Image, im2: Image.Image, sigmas: float):
    dst = Image.new('RGB', (im1.width + im2.width, im1.height))
    dst.paste(im1.convert('RGB'), (0, 0))
    dst.paste(im2.convert('RGB'), (im1.width, 0))
    font = ImageFont.truetype('Roboto-Regular.ttf', 72, encoding='unic')
    draw = ImageDraw.Draw(dst)
    draw.text((64, im1.height - 128), 'Default Flux', 'red', font=font)
    draw.text((im1.width + 64, im1.height - 128), f'Sigmas * factor {sigmas}', 'red', font=font)
    return dst

# @spaces.GPU(duration=90)
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, mul_sigmas=0.95, is_cmp=True, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)
    
    sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
    sigmas = sigmas * mul_sigmas

    image_sigmas = pipe(
            prompt=prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
            output_type="pil",
            sigmas=sigmas
        ).images[0]
    
    if is_cmp:
        image_def = pipe(
                prompt=prompt,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps,
                width=width,
                height=height,
                generator=generator,
                output_type="pil",
            ).images[0]
        return [image_def, image_sigmas], get_cmp_image(image_def, image_sigmas, mul_sigmas), seed
    else: return [image_sigmas, image_sigmas], None, seed

examples = [
    "a tiny astronaut hatching from an egg on the moon",
    "a cat holding a sign that says hello world",
    "an anime illustration of a wiener schnitzel",
]

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"""# FLUX.1 [dev] sigmas test
12B param rectified flow transformer guidance-distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/)  
[[non-commercial license](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)] [[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-dev)]
        """)
        
        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)
        result = ImageSlider(label="Result", show_label=False, type="pil", slider_color="pink")
        result_cmp = gr.Image(label="Result (comparing)", show_label=False, type="pil", format="png", height=256, show_download_button=True, show_share_button=False)
        
        with gr.Accordion("Advanced Settings", open=True):
            with gr.Row():
                sigmas = gr.Slider(
                    label="Sigmas",
                    minimum=0,
                    maximum=1.0,
                    step=0.01,
                    value=0.95,
                )
                is_cmp = gr.Checkbox(label="Compare images with/without sigmas", value=True)
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=9119,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=False)
            
            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=1,
                    maximum=15,
                    step=0.1,
                    value=3.5,
                )
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )

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

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

demo.launch()