import gradio as gr
from datasets import load_dataset
import jax
import numpy as np
import jax.numpy as jnp
from flax.jax_utils import replicate
from flax.training.common_utils import shard
#from diffusers.utils import load_image
from diffusers.utils.testing_utils import load_image
from PIL import Image
from diffusers import FlaxStableDiffusionControlNetPipeline, FlaxControlNetModel

def image_grid(imgs, rows, cols):
    w, h = imgs[0].size
    grid = Image.new("RGB", size=(cols * w, rows * h))
    for i, img in enumerate(imgs):
        grid.paste(img, box=(i % cols * w, i // cols * h))
    return grid

def create_key(seed=0):
    return jax.random.PRNGKey(seed)


def infer(prompt, negative_prompt, image):
    rng = create_key(0)

#    canny_image = load_image(
#        "https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/blog_post_cell_10_output_0.jpeg"
#    )
    canny_image = load_image(image)

    #prompts = "a living room fan"
    prompts = prompt
    negative_prompts = negative_prompt

    controlnet, controlnet_params = FlaxControlNetModel.from_pretrained(
        "tsungtao/controlnet-mlsd-202305011046", from_flax=True, dtype=jnp.float32
    )

    pipe, params = FlaxStableDiffusionControlNetPipeline.from_pretrained(
        "runwayml/stable-diffusion-v1-5", controlnet=controlnet, revision="flax", dtype=jnp.float32
    )

    params["controlnet"] = controlnet_params

    num_samples = jax.device_count()
    rng = jax.random.split(rng, jax.device_count())

    prompt_ids = pipe.prepare_text_inputs([prompts] * num_samples)
    negative_prompt_ids = pipe.prepare_text_inputs([negative_prompts] * num_samples)
    processed_image = pipe.prepare_image_inputs([canny_image] * num_samples)

    p_params = replicate(params)
    prompt_ids = shard(prompt_ids)
    negative_prompt_ids = shard(negative_prompt_ids)
    processed_image = shard(processed_image)

    output = pipe(
        prompt_ids=prompt_ids,
        image=processed_image,
        params=p_params,
        prng_seed=rng,
        num_inference_steps=50,
        neg_prompt_ids=negative_prompt_ids,
        jit=True,
    ).images

    output_images = pipe.numpy_to_pil(np.asarray(output.reshape((num_samples,) + output.shape[-3:])))
    output_images = image_grid(output_images, num_samples // 4, 4)
    #output_images.save("tao/image.png")
    #dataset = load_dataset('imagefolder', data_dir='tao')
    #dataset.push_to_hub('tsungtao/tmp')
    return output_images

#infer('','','')

def infer2(prompt, negative_prompt, image):
    output_image = infer(prompt, negative_prompt, image)
    #output_image = "https://datasets-server.huggingface.co/assets/tsungtao/tmp/--/tsungtao--tmp/train/0/image/image.jpg"
    return output_image

title = "ControlNet on MLSD Filter"
description = "This is a demo on ControlNet based on mlsd filter."

#examples = [["living room with TV", "fan", "https://datasets-server.huggingface.co/assets/tsungtao/diffusers-testing/--/tsungtao--diffusers-testing/train/0/images/image.jpg"]]

interface = gr.Interface(fn = infer2, inputs = ["text", "text", "text"], outputs = "image",title = title, description = description, theme='gradio/soft')
interface.launch(enable_queue=True)