from test import *

import copy
import gradio as gr

class merge_config_then_run():
    def __init__(self) -> None:
            # Load the tokenizer
        pretrained_model_path = '/home/xianyang/Data/code/FateZero/ckpt/stable-diffusion-v1-5'
        self.tokenizer = None
        self.text_encoder = None
        self.vae = None
        self.unet = None
        
        cache_ckpt = True
        if cache_ckpt:
            self.tokenizer = AutoTokenizer.from_pretrained(
                pretrained_model_path,
                # 'FateZero/ckpt/stable-diffusion-v1-4',
                subfolder="tokenizer",
                use_fast=False,
            )

            # Load models and create wrapper for stable diffusion
            self.text_encoder = CLIPTextModel.from_pretrained(
                pretrained_model_path,
                subfolder="text_encoder",
            )

            self.vae = AutoencoderKL.from_pretrained(
                pretrained_model_path,
                subfolder="vae",
            )

            self.unet = UNetPseudo3DConditionModel.from_2d_model(
                os.path.join(pretrained_model_path, "unet"), model_config=model_config
            )

    def run(
        self,
        # def merge_config_then_run(
        model_id,
        data_path,
        source_prompt,
        target_prompt,
        cross_replace_steps,
        self_replace_steps,
        enhance_words,
        enhance_words_value,
        num_steps,
        guidance_scale,
        user_input_video=None,
        
        # Temporal and spatial crop of the video
        start_sample_frame=0,
        n_sample_frame=8,
        stride=1,
        left_crop=0,
        right_crop=0,
        top_crop=0,
        bottom_crop=0,
    ):
        # , ] = inputs
        default_edit_config='config/demo_config.yaml'
        Omegadict_default_edit_config = OmegaConf.load(default_edit_config)
        
        dataset_time_string = get_time_string()
        config_now = copy.deepcopy(Omegadict_default_edit_config)
        print(f"config_now['pretrained_model_path'] = model_id {model_id}")
        # config_now['pretrained_model_path'] = model_id
        config_now['dataset_config']['prompt'] = source_prompt
        config_now['dataset_config']['path'] = data_path
        # ImageSequenceDataset_dict = { }
        offset_dict = {
            "left": left_crop,
            "right": right_crop,
            "top": top_crop,
            "bottom": bottom_crop,
        }
        ImageSequenceDataset_dict = {
            "start_sample_frame" : start_sample_frame,
            "n_sample_frame" : n_sample_frame,
            "sampling_rate"       : stride,   
            "offset": offset_dict,
        }
        config_now['dataset_config'].update(ImageSequenceDataset_dict)
        if user_input_video and data_path is None:
            raise gr.Error('You need to upload a video or choose a provided video')
        if user_input_video is not None:
            if isinstance(user_input_video, str):
                config_now['dataset_config']['path'] = user_input_video
            elif hasattr(user_input_video, 'name') and user_input_video.name is not None:
                config_now['dataset_config']['path'] = user_input_video.name


        # editing config
        config_now['editing_config']['prompts'] = [target_prompt]

        config_now['editing_config']['guidance_scale'] = guidance_scale
        config_now['editing_config']['num_inference_steps'] = num_steps
        

        logdir = default_edit_config.replace('config', 'result').replace('.yml', '').replace('.yaml', '')+f'_{dataset_time_string}'      
        config_now['logdir'] = logdir
        print(f'Saving at {logdir}')
        save_path = test(tokenizer = self.tokenizer,
                         text_encoder = self.text_encoder,
                         vae = self.vae,
                         unet = self.unet,
                         config=default_edit_config, **config_now)
        mp4_path = save_path.replace('_0.gif', '_0_0_0.mp4')
        return mp4_path