app.py

import gradio as gr
import torch
from diffusers import FluxPipeline, StableDiffusion3Pipeline
from PIL import Image

import random
import numpy as np
import spaces

from FlowEdit_utils import FlowEditSD3, FlowEditFLUX


device = "cuda" if torch.cuda.is_available() else "cpu"
# device = "cpu"
# model_type = 'SD3'

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
scheduler = pipe.scheduler
pipe = pipe.to(device)
loaded_model = 'SD3'


def on_model_change(model_type):
    if model_type == 'SD3':

        T_steps_value = 50

        src_guidance_scale_value = 3.5

        tar_guidance_scale_value = 13.5

        n_max_value = 33

    elif model_type == 'FLUX':

        T_steps_value = 28

        src_guidance_scale_value = 1.5

        tar_guidance_scale_value = 5.5

        n_max_value = 24

    else:
        raise NotImplementedError(f"Model type {model_type} not implemented")
    
    return T_steps_value, src_guidance_scale_value, tar_guidance_scale_value, n_max_value

def get_examples():
    case = [
        ["inputs/cat.png", "SD3", 50,  3.5, 13.5, 33, "a cat sitting in the grass", "a puppy sitting in the grass", 0, 1, 42],
        ["inputs/gas_station.png", "SD3", 50,  3.5, 13.5, 33, "cars are parked in front of a gas station with a sign that says \"CAFE\"", "cars are parked in front of a gas station with a sign that says \"CVPR\"", 0, 1, 42],
        ["inputs/iguana.png", "SD3", 50,  3.5, 13.5, 31, "A large orange lizard sitting on a rock near the      ocean. The lizard is positioned in the center of the scene, with the ocean waves visible in the background. The rock is located close to the water, providing a picturesque setting for the lizard''s resting spot.", "A large dragon sitting on a rock near the ocean. The dragon is positioned in the center of the scene, with the ocean waves visible in the background. The rock is located close to the water, providing a picturesque setting for the dragon''s resting spot.", 0, 1, 42],
        ["inputs/cat.png", "FLUX", 28,  1.5, 5.5, 24, "a cat sitting in the grass", "a puppy sitting in the grass", 0, 1, 42],
        ["inputs/gas_station.png", "FLUX", 28,  1.5, 5.5, 24, "cars are parked in front of a gas station with a sign that says \"CAFE\"", "cars are parked in front of a gas station with a sign that says \"CVPR\"", 0, 1, 23],
        ["inputs/steak.png", "FLUX", 28,  1.5, 5.5, 24, "A steak accompanied by a side of leaf salad.", "A bread roll accompanied by a side of leaf salad.", 0, 1, 42],
    ]
    return case


@spaces.GPU()
def FlowEditRun(
    image_src: str,
    model_type: str,
    T_steps: int,
    src_guidance_scale: float,
    tar_guidance_scale: float,
    n_max: int,
    src_prompt: str,
    tar_prompt: str,
    n_min: int,
    n_avg: int,
    seed: int,

    ):

    if not len(src_prompt):
        raise gr.Error("source prompt cannot be empty")
    if not len(tar_prompt):
        raise gr.Error("target prompt cannot be empty")

    global pipe
    global scheduler
    global loaded_model

    # reload model only if different from the loaded model
    if loaded_model != model_type:

        if model_type == 'FLUX':
            # pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.float16) 
            pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.float16)
            loaded_model = 'FLUX'
        elif model_type == 'SD3':
            pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
            loaded_model = 'SD3'
        else:
            raise NotImplementedError(f"Model type {model_type} not implemented")

        scheduler = pipe.scheduler
        pipe = pipe.to(device)




    # set seed
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    # load image
    image = Image.open(image_src)
    # crop image to have both dimensions divisibe by 16 - avoids issues with resizing
    image = image.crop((0, 0, image.width - image.width % 16, image.height - image.height % 16))
    image_src = pipe.image_processor.preprocess(image)
    # image_tar = pipe.image_processor.postprocess(image_src)
    # return image_tar[0]

    # cast image to half precision
    image_src = image_src.to(device).half()

    with torch.autocast("cuda"), torch.inference_mode():
        x0_src_denorm = pipe.vae.encode(image_src).latent_dist.mode()
    x0_src = (x0_src_denorm - pipe.vae.config.shift_factor) * pipe.vae.config.scaling_factor
    # send to cuda
    x0_src = x0_src.to(device)

    negative_prompt =  "" # optionally add support for negative prompts (SD3)

    if model_type == 'SD3':
        x0_tar = FlowEditSD3(pipe,
                            scheduler,
                            x0_src,
                            src_prompt,
                            tar_prompt,
                            negative_prompt,
                            T_steps,
                            n_avg,
                            src_guidance_scale,
                            tar_guidance_scale,
                            n_min,
                            n_max,)
        
    elif model_type == 'FLUX':
        x0_tar = FlowEditFLUX(pipe,
                            scheduler,
                            x0_src,
                            src_prompt,
                            tar_prompt,
                            negative_prompt,
                            T_steps,
                            n_avg,
                            src_guidance_scale,
                            tar_guidance_scale,
                            n_min,
                            n_max,)
    else:
        raise NotImplementedError(f"Sampler type {model_type} not implemented")


    x0_tar_denorm = (x0_tar / pipe.vae.config.scaling_factor) + pipe.vae.config.shift_factor
    with torch.autocast("cuda"), torch.inference_mode():
        image_tar = pipe.vae.decode(x0_tar_denorm, return_dict=False)[0]
    image_tar = pipe.image_processor.postprocess(image_tar)


    return image_tar[0]


# title = "FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models"

intro = """
<h1 style="font-weight: 1000; text-align: center; margin: 0px;">FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models</h1>
<h3 style="margin-bottom: 10px; text-align: center;">
    <a href="https://arxiv.org/">[Paper]</a>&nbsp;|&nbsp;
    <a href="https://matankleiner.github.io/flowedit/">[Project Page]</a>&nbsp;|&nbsp;
    <a href="https://github.com/fallenshock/FlowEdit">[Code]</a>
</h3>
Gradio demo for FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models. See our project page for more details.

<br>
<br>Edit your image using Flow models! upload an image, add a description of it, and specify the edits you want to make.
<h3>Notes:</h3>

<ol>
  <li>We use FLUX.1 dev and SD3 for the demo. The models are large and may take a while to load.</li>
  <li>We recommend 1024x1024 images for the best results. If the input images are too large, there may be out-of-memory errors.</li>
  <li>Default hyperparameters for each model used in the paper are provided as examples. Feel free to experiment with them as well.</li>
</ol>  

"""

# article = """
# 📝 **Citation**
# ```bibtex
# @article{aaa,
#   author    = {},
#   title     = {},
#   journal   = {},
#   year      = {2024},
#   url       = {}
# }
# ```
# """


with gr.Blocks() as demo:
    

    gr.HTML(intro)

    with gr.Row(equal_height=True):
        image_src = gr.Image(type="filepath", label="Source Image", value="inputs/cat.png",)
        image_tar = gr.Image(label="Output", type="pil", show_label=True, format="png",),

    with gr.Row():
        src_prompt = gr.Textbox(lines=2, label="Source Prompt", value="a cat sitting in the grass")

    with gr.Row():
        tar_prompt = gr.Textbox(lines=2, label="Target Prompt", value="a puppy sitting in the grass")

    with gr.Row():
        model_type = gr.Dropdown(["SD3", "FLUX"], label="Model Type", value="SD3")
        T_steps = gr.Number(value=50, label="Total Steps", minimum=1, maximum=50)
        n_max = gr.Number(value=33, label="n_max (control the strength of the edit)")

    with gr.Row():
        src_guidance_scale = gr.Slider(minimum=1.0, maximum=30.0, value=3.5, label="src_guidance_scale")
        tar_guidance_scale = gr.Slider(minimum=1.0, maximum=30.0, value=13.5, label="tar_guidance_scale")
    
    with gr.Row():
        submit_button = gr.Button("Run FlowEdit", variant="primary")

    with gr.Accordion(label="Advanced Settings", open=False):
        # additional inputs
        n_min = gr.Number(value=0, label="n_min (for improved style edits)")
        n_avg = gr.Number(value=1, label="n_avg (improve structure at the cost of runtime)", minimum=1)
        seed = gr.Number(value=42, label="seed")



    submit_button.click(
                        fn=FlowEditRun, 
                        inputs=[
                        image_src,
                        model_type,
                        T_steps,
                        src_guidance_scale,
                        tar_guidance_scale,
                        n_max,
                        src_prompt,
                        tar_prompt,
                        n_min,
                        n_avg,
                        seed,
                        ],
                        outputs=[
                        image_tar[0],
                        ],
                        )

    gr.Examples(
        label="Examples",
        examples=get_examples(),
        inputs=[image_src, model_type, T_steps, src_guidance_scale, tar_guidance_scale, n_max, src_prompt, tar_prompt, n_min, n_avg, seed],
    )

    model_type.input(fn=on_model_change, inputs=[model_type], outputs=[T_steps, src_guidance_scale, tar_guidance_scale, n_max])


    # gr.HTML(article)
demo.queue()
demo.launch( )