facefusion/processors/modules/face_swapper.py

from argparse import ArgumentParser
from functools import lru_cache
from typing import List, Tuple

import os
import tempfile
from facefusion import wording  # Если wording не импортирован
import cv2
import numpy

import facefusion.choices
import facefusion.jobs.job_manager
import facefusion.jobs.job_store
import facefusion.processors.core as processors
from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, wording
from facefusion.common_helper import get_first
from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
from facefusion.execution import has_execution_provider
from facefusion.face_analyser import get_average_face, get_many_faces, get_one_face
from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
from facefusion.face_selector import find_similar_faces, sort_and_filter_faces, sort_faces_by_order
from facefusion.face_store import get_reference_faces
from facefusion.filesystem import filter_image_paths, has_image, in_directory, is_image, is_video, resolve_relative_path, same_file_extension
from facefusion.model_helper import get_static_model_initializer
from facefusion.processors import choices as processors_choices
from facefusion.processors.pixel_boost import explode_pixel_boost, implode_pixel_boost
from facefusion.processors.types import FaceSwapperInputs
from facefusion.program_helper import find_argument_group
from facefusion.thread_helper import conditional_thread_semaphore
from facefusion.types import ApplyStateItem, Args, DownloadScope, Embedding, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
from facefusion.vision import read_image, read_static_image, read_static_images, unpack_resolution, write_image


@lru_cache(maxsize = None)
def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
	return\
	{
		'blendswap_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'blendswap_256.hash'),
					'path': resolve_relative_path('../.assets/models/blendswap_256.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'blendswap_256.onnx'),
					'path': resolve_relative_path('../.assets/models/blendswap_256.onnx')
				}
			},
			'type': 'blendswap',
			'template': 'ffhq_512',
			'size': (256, 256),
			'mean': [ 0.0, 0.0, 0.0 ],
			'standard_deviation': [ 1.0, 1.0, 1.0 ]
		},
		'ghost_1_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_1_256.hash'),
					'path': resolve_relative_path('../.assets/models/ghost_1_256.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_1_256.onnx'),
					'path': resolve_relative_path('../.assets/models/ghost_1_256.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
				}
			},
			'type': 'ghost',
			'template': 'arcface_112_v1',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'ghost_2_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_2_256.hash'),
					'path': resolve_relative_path('../.assets/models/ghost_2_256.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_2_256.onnx'),
					'path': resolve_relative_path('../.assets/models/ghost_2_256.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
				}
			},
			'type': 'ghost',
			'template': 'arcface_112_v1',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'ghost_3_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_3_256.hash'),
					'path': resolve_relative_path('../.assets/models/ghost_3_256.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'ghost_3_256.onnx'),
					'path': resolve_relative_path('../.assets/models/ghost_3_256.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_ghost.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
				}
			},
			'type': 'ghost',
			'template': 'arcface_112_v1',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'hififace_unofficial_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.1.0', 'hififace_unofficial_256.hash'),
					'path': resolve_relative_path('../.assets/models/hififace_unofficial_256.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.1.0', 'arcface_converter_hififace.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_hififace.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.1.0', 'hififace_unofficial_256.onnx'),
					'path': resolve_relative_path('../.assets/models/hififace_unofficial_256.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.1.0', 'arcface_converter_hififace.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_hififace.onnx')
				}
			},
			'type': 'hififace',
			'template': 'mtcnn_512',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'hyperswap_1a_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.3.0', 'hyperswap_1a_256.hash'),
					'path': resolve_relative_path('../.assets/models/hyperswap_1a_256.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.3.0', 'hyperswap_1a_256.onnx'),
					'path': resolve_relative_path('../.assets/models/hyperswap_1a_256.onnx')
				}
			},
			'type': 'hyperswap',
			'template': 'arcface_128',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'hyperswap_1b_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.3.0', 'hyperswap_1b_256.hash'),
					'path': resolve_relative_path('../.assets/models/hyperswap_1b_256.hash')
				}
			},
			'sources':
				{
					'face_swapper':
					{
						'url': resolve_download_url('models-3.3.0', 'hyperswap_1b_256.onnx'),
						'path': resolve_relative_path('../.assets/models/hyperswap_1b_256.onnx')
					}
				},
			'type': 'hyperswap',
			'template': 'arcface_128',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'hyperswap_1c_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.3.0', 'hyperswap_1c_256.hash'),
					'path': resolve_relative_path('../.assets/models/hyperswap_1c_256.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.3.0', 'hyperswap_1c_256.onnx'),
					'path': resolve_relative_path('../.assets/models/hyperswap_1c_256.onnx')
				}
			},
			'type': 'hyperswap',
			'template': 'arcface_128',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		},
		'inswapper_128':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'inswapper_128.hash'),
					'path': resolve_relative_path('../.assets/models/inswapper_128.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'inswapper_128.onnx'),
					'path': resolve_relative_path('../.assets/models/inswapper_128.onnx')
				}
			},
			'type': 'inswapper',
			'template': 'arcface_128',
			'size': (128, 128),
			'mean': [ 0.0, 0.0, 0.0 ],
			'standard_deviation': [ 1.0, 1.0, 1.0 ]
		},
		'inswapper_128_fp16':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'inswapper_128_fp16.hash'),
					'path': resolve_relative_path('../.assets/models/inswapper_128_fp16.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'inswapper_128_fp16.onnx'),
					'path': resolve_relative_path('../.assets/models/inswapper_128_fp16.onnx')
				}
			},
			'type': 'inswapper',
			'template': 'arcface_128',
			'size': (128, 128),
			'mean': [ 0.0, 0.0, 0.0 ],
			'standard_deviation': [ 1.0, 1.0, 1.0 ]
		},
		'simswap_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'simswap_256.hash'),
					'path': resolve_relative_path('../.assets/models/simswap_256.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_simswap.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'simswap_256.onnx'),
					'path': resolve_relative_path('../.assets/models/simswap_256.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_simswap.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.onnx')
				}
			},
			'type': 'simswap',
			'template': 'arcface_112_v1',
			'size': (256, 256),
			'mean': [ 0.485, 0.456, 0.406 ],
			'standard_deviation': [ 0.229, 0.224, 0.225 ]
		},
		'simswap_unofficial_512':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'simswap_unofficial_512.hash'),
					'path': resolve_relative_path('../.assets/models/simswap_unofficial_512.hash')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_simswap.hash'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'simswap_unofficial_512.onnx'),
					'path': resolve_relative_path('../.assets/models/simswap_unofficial_512.onnx')
				},
				'embedding_converter':
				{
					'url': resolve_download_url('models-3.0.0', 'arcface_converter_simswap.onnx'),
					'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.onnx')
				}
			},
			'type': 'simswap',
			'template': 'arcface_112_v1',
			'size': (512, 512),
			'mean': [ 0.0, 0.0, 0.0 ],
			'standard_deviation': [ 1.0, 1.0, 1.0 ]
		},
		'uniface_256':
		{
			'hashes':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'uniface_256.hash'),
					'path': resolve_relative_path('../.assets/models/uniface_256.hash')
				}
			},
			'sources':
			{
				'face_swapper':
				{
					'url': resolve_download_url('models-3.0.0', 'uniface_256.onnx'),
					'path': resolve_relative_path('../.assets/models/uniface_256.onnx')
				}
			},
			'type': 'uniface',
			'template': 'ffhq_512',
			'size': (256, 256),
			'mean': [ 0.5, 0.5, 0.5 ],
			'standard_deviation': [ 0.5, 0.5, 0.5 ]
		}
	}


def get_inference_pool() -> InferencePool:
	model_names = [ get_model_name() ]
	model_source_set = get_model_options().get('sources')

	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)


def clear_inference_pool() -> None:
	model_names = [ get_model_name() ]
	inference_manager.clear_inference_pool(__name__, model_names)


def get_model_options() -> ModelOptions:
	model_name = get_model_name()
	return create_static_model_set('full').get(model_name)


def get_model_name() -> str:
	model_name = state_manager.get_item('face_swapper_model')

	if has_execution_provider('coreml') and model_name == 'inswapper_128_fp16':
		return 'inswapper_128'
	return model_name


def register_args(program : ArgumentParser) -> None:
	group_processors = find_argument_group(program, 'processors')
	if group_processors:
		group_processors.add_argument('--face-swapper-model', help = wording.get('help.face_swapper_model'), default = config.get_str_value('processors', 'face_swapper_model', 'inswapper_128_fp16'), choices = processors_choices.face_swapper_models)
		known_args, _ = program.parse_known_args()
		face_swapper_pixel_boost_choices = processors_choices.face_swapper_set.get(known_args.face_swapper_model)
		group_processors.add_argument('--face-swapper-pixel-boost', help = wording.get('help.face_swapper_pixel_boost'), default = config.get_str_value('processors', 'face_swapper_pixel_boost', get_first(face_swapper_pixel_boost_choices)), choices = face_swapper_pixel_boost_choices)
		facefusion.jobs.job_store.register_step_keys([ 'face_swapper_model', 'face_swapper_pixel_boost' ])


def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
	apply_state_item('face_swapper_model', args.get('face_swapper_model'))
	apply_state_item('face_swapper_pixel_boost', args.get('face_swapper_pixel_boost'))


def pre_check() -> bool:
	model_hash_set = get_model_options().get('hashes')
	model_source_set = get_model_options().get('sources')

	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


def pre_process(mode : ProcessMode) -> bool:
	if not has_image(state_manager.get_item('source_paths')):
		logger.error(wording.get('choose_image_source') + wording.get('exclamation_mark'), __name__)
		return False
	source_image_paths = filter_image_paths(state_manager.get_item('source_paths'))
	source_frames = read_static_images(source_image_paths)
	source_faces = get_many_faces(source_frames)
	if not get_one_face(source_faces):
		logger.error(wording.get('no_source_face_detected') + wording.get('exclamation_mark'), __name__)
		return False
	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
		return False
	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
		return False
	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
		return False
	return True


def post_process() -> None:
	read_static_image.cache_clear()
	video_manager.clear_video_pool()
	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
		get_static_model_initializer.cache_clear()
		clear_inference_pool()
	if state_manager.get_item('video_memory_strategy') == 'strict':
		content_analyser.clear_inference_pool()
		face_classifier.clear_inference_pool()
		face_detector.clear_inference_pool()
		face_landmarker.clear_inference_pool()
		face_masker.clear_inference_pool()
		face_recognizer.clear_inference_pool()


def swap_face(source_face: Face, target_face: Face, temp_vision_frame: VisionFrame) -> VisionFrame:
#   print(f"[DEBUG] Starting face swap")
#   print(f"[DEBUG] Source face landmarks: {source_face.landmark_set.get('5/68').shape if source_face.landmark_set.get('5/68') is not None else 'None'}")
#   print(f"[DEBUG] Target face landmarks: {target_face.landmark_set.get('5/68').shape if target_face.landmark_set.get('5/68') is not None else 'None'}")
    
    model_template = get_model_options().get('template')
    model_size = get_model_options().get('size')
    pixel_boost_size = unpack_resolution(state_manager.get_item('face_swapper_pixel_boost'))
    pixel_boost_total = pixel_boost_size[0] // model_size[0]
    
#   print(f"[DEBUG] Model template: {model_template}")
#   print(f"[DEBUG] Model size: {model_size}")
#   print(f"[DEBUG] Pixel boost size: {pixel_boost_size}")
    
    try:
        crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(
            temp_vision_frame, 
            target_face.landmark_set.get('5/68'),
            model_template,
            pixel_boost_size
        )
    #   print(f"[DEBUG] Warped face shape: {crop_vision_frame.shape}")
    except Exception as e:
    #   print(f"[DEBUG] Error in face warping: {e}")
        return temp_vision_frame
    
    temp_vision_frames = []
    crop_masks = []

    if 'box' in state_manager.get_item('face_mask_types'):
    #   print("[DEBUG] Creating box mask")
        box_mask = create_box_mask(crop_vision_frame, state_manager.get_item('face_mask_blur'), state_manager.get_item('face_mask_padding'))
        crop_masks.append(box_mask)

    if 'occlusion' in state_manager.get_item('face_mask_types'):
    #   print("[DEBUG] Creating occlusion mask")
        occlusion_mask = create_occlusion_mask(crop_vision_frame)
        crop_masks.append(occlusion_mask)

#   print("[DEBUG] Starting pixel boost processing")
    pixel_boost_vision_frames = implode_pixel_boost(crop_vision_frame, pixel_boost_total, model_size)
#   print(f"[DEBUG] Created {len(pixel_boost_vision_frames)} pixel boost frames")
    
    for idx, pixel_boost_vision_frame in enumerate(pixel_boost_vision_frames):
    #   print(f"[DEBUG] Processing pixel boost frame {idx + 1}/{len(pixel_boost_vision_frames)}")
        try:
            pixel_boost_vision_frame = prepare_crop_frame(pixel_boost_vision_frame)
            pixel_boost_vision_frame = forward_swap_face(source_face, pixel_boost_vision_frame)
            pixel_boost_vision_frame = normalize_crop_frame(pixel_boost_vision_frame)
            temp_vision_frames.append(pixel_boost_vision_frame)
        except Exception as e:
        #   print(f"[DEBUG] Error processing pixel boost frame: {e}")
            return temp_vision_frame
            
#   print("[DEBUG] Exploding pixel boost frames")
    crop_vision_frame = explode_pixel_boost(temp_vision_frames, pixel_boost_total, model_size, pixel_boost_size)

    if 'area' in state_manager.get_item('face_mask_types'):
    #   print("[DEBUG] Creating area mask")
        try:
            face_landmark_68 = cv2.transform(target_face.landmark_set.get('68').reshape(1, -1, 2), affine_matrix).reshape(-1, 2)
            area_mask = create_area_mask(crop_vision_frame, face_landmark_68, state_manager.get_item('face_mask_areas'))
            crop_masks.append(area_mask)
        except Exception as e:
            print(f"[DEBUG] Error creating area mask: {e}")

    if 'region' in state_manager.get_item('face_mask_types'):
    #   print("[DEBUG] Creating region mask")
        region_mask = create_region_mask(crop_vision_frame, state_manager.get_item('face_mask_regions'))
        crop_masks.append(region_mask)

#   print(f"[DEBUG] Created {len(crop_masks)} masks")
    crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
#   print("[DEBUG] Pasting back face")
    temp_vision_frame = paste_back(temp_vision_frame, crop_vision_frame, crop_mask, affine_matrix)
#   print("[DEBUG] Face swap completed")
    return temp_vision_frame


def forward_swap_face(source_face: Face, crop_vision_frame: VisionFrame) -> VisionFrame:
#   print("[DEBUG] Starting forward face swap")
    face_swapper = get_inference_pool().get('face_swapper')
    model_type = get_model_options().get('type')
    face_swapper_inputs = {}

#   print(f"[DEBUG] Model type: {model_type}")

    if has_execution_provider('coreml') and model_type in ['ghost', 'uniface']:
        face_swapper.set_providers([facefusion.choices.execution_provider_set.get('cpu')])

    try:
        for face_swapper_input in face_swapper.get_inputs():
            if face_swapper_input.name == 'source':
                if model_type in ['blendswap', 'uniface']:
                    face_swapper_inputs[face_swapper_input.name] = prepare_source_frame(source_face)
                else:
                    face_swapper_inputs[face_swapper_input.name] = prepare_source_embedding(source_face)
            #   print(f"[DEBUG] Prepared source input: {face_swapper_input.name}")
            if face_swapper_input.name == 'target':
                face_swapper_inputs[face_swapper_input.name] = crop_vision_frame
            #   print(f"[DEBUG] Prepared target input: {face_swapper_input.name}")

    #   print("[DEBUG] Running face swapper")
        with conditional_thread_semaphore():
            result = face_swapper.run(None, face_swapper_inputs)[0][0]
    #   print("[DEBUG] Face swap completed successfully")
        return result
    except Exception as e:
    #   print(f"[DEBUG] Error in forward face swap: {e}")
        return crop_vision_frame


def forward_convert_embedding(embedding : Embedding) -> Embedding:
	embedding_converter = get_inference_pool().get('embedding_converter')

	with conditional_thread_semaphore():
		embedding = embedding_converter.run(None,
		{
			'input': embedding
		})[0]

	return embedding


def prepare_source_frame(source_face : Face) -> VisionFrame:
	model_type = get_model_options().get('type')
	source_vision_frame = read_static_image(get_first(state_manager.get_item('source_paths')))

	if model_type == 'blendswap':
		source_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, source_face.landmark_set.get('5/68'), 'arcface_112_v2', (112, 112))
	if model_type == 'uniface':
		source_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, source_face.landmark_set.get('5/68'), 'ffhq_512', (256, 256))
	source_vision_frame = source_vision_frame[:, :, ::-1] / 255.0
	source_vision_frame = source_vision_frame.transpose(2, 0, 1)
	source_vision_frame = numpy.expand_dims(source_vision_frame, axis = 0).astype(numpy.float32)
	return source_vision_frame


def prepare_source_embedding(source_face: Face) -> Embedding:
#   print("[DEBUG] Preparing source embedding")
    model_type = get_model_options().get('type')

    try:
        if model_type == 'ghost':
            source_embedding, _ = convert_embedding(source_face)
            source_embedding = source_embedding.reshape(1, -1)
        #   print("[DEBUG] Prepared ghost embedding")
            return source_embedding

        if model_type == 'hyperswap':
            source_embedding = source_face.normed_embedding.reshape((1, -1))
        #   print("[DEBUG] Prepared hyperswap embedding")
            return source_embedding

        if model_type == 'inswapper':
            model_path = get_model_options().get('sources').get('face_swapper').get('path')
            model_initializer = get_static_model_initializer(model_path)
            source_embedding = source_face.embedding.reshape((1, -1))
            source_embedding = numpy.dot(source_embedding, model_initializer) / numpy.linalg.norm(source_embedding)
        #   print("[DEBUG] Prepared inswapper embedding")
            return source_embedding

        _, source_normed_embedding = convert_embedding(source_face)
        source_embedding = source_normed_embedding.reshape(1, -1)
    #   print("[DEBUG] Prepared default embedding")
        return source_embedding
    except Exception as e:
    #   print(f"[DEBUG] Error preparing source embedding: {e}")
        raise


def convert_embedding(source_face : Face) -> Tuple[Embedding, Embedding]:
	embedding = source_face.embedding.reshape(-1, 512)
	embedding = forward_convert_embedding(embedding)
	embedding = embedding.ravel()
	normed_embedding = embedding / numpy.linalg.norm(embedding)
	return embedding, normed_embedding


def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
	model_mean = get_model_options().get('mean')
	model_standard_deviation = get_model_options().get('standard_deviation')

	crop_vision_frame = crop_vision_frame[:, :, ::-1] / 255.0
	crop_vision_frame = (crop_vision_frame - model_mean) / model_standard_deviation
	crop_vision_frame = crop_vision_frame.transpose(2, 0, 1)
	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0).astype(numpy.float32)
	return crop_vision_frame


def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
	model_type = get_model_options().get('type')
	model_mean = get_model_options().get('mean')
	model_standard_deviation = get_model_options().get('standard_deviation')

	crop_vision_frame = crop_vision_frame.transpose(1, 2, 0)
	if model_type in [ 'ghost', 'hififace', 'hyperswap', 'uniface' ]:
		crop_vision_frame = crop_vision_frame * model_standard_deviation + model_mean
	crop_vision_frame = crop_vision_frame.clip(0, 1)
	crop_vision_frame = crop_vision_frame[:, :, ::-1] * 255
	return crop_vision_frame


def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
	return swap_face(source_face, target_face, temp_vision_frame)


def process_frame(inputs: FaceSwapperInputs) -> VisionFrame:
    try:
        reference_faces = inputs.get('reference_faces')
        source_face = inputs.get('source_face')
        target_vision_frame = inputs.get('target_vision_frame')
        is_preview = inputs.get('preview', False)
        
    #   print(f"[DEBUG] Processing frame for {'preview' if is_preview else 'final output'}")
    #   print(f"[DEBUG] Source face present: {source_face is not None}")
    #   print(f"[DEBUG] Reference faces present: {reference_faces is not None}")
    #   print(f"[DEBUG] Target frame shape: {target_vision_frame.shape if target_vision_frame is not None else 'None'}")
        
        many_faces = sort_and_filter_faces(get_many_faces([target_vision_frame]))
    #   print(f"[DEBUG] Found {len(many_faces) if many_faces else 0} faces in target frame")
        
        if not many_faces:
        #   print("[DEBUG] No faces found — returning original frame")
            return target_vision_frame
        
        face_selector_mode = state_manager.get_item('face_selector_mode')
        face_selector_order = state_manager.get_item('face_selector_order')
    #   print(f"[DEBUG] face_selector_mode: '{face_selector_mode}', order: '{face_selector_order}'")
        
        if face_selector_mode == 'reference' and reference_faces:
        #   print("[DEBUG] Entering reference mode")
            similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
        #   print(f"[DEBUG] Found {len(similar_faces) if similar_faces else 0} similar faces")
            if similar_faces:
                result_frame = target_vision_frame.copy()
                for similar_face in similar_faces:
                    result_frame = swap_face(source_face, similar_face, result_frame)
            #   print(f"[DEBUG] Face swap completed in reference mode, shape: {result_frame.shape}")
                return result_frame
        elif face_selector_mode == 'one':
        #   print("[DEBUG] Entering 'one' mode")
            try:
                selected_faces = sort_faces_by_order(many_faces, face_selector_order)
            #   print(f"[DEBUG] Sorted faces: {len(selected_faces) if selected_faces else 0}")
                
                # Фикс: Брать face_selector_index из inputs (не из state_manager)
                face_index_raw = inputs.get('face_selector_index')
                face_index = int(face_index_raw) if face_index_raw is not None and str(face_index_raw).isdigit() else 0
            #   print(f"[DEBUG] face_index from inputs: {face_index} (raw: {face_index_raw}), total selected: {len(selected_faces) if selected_faces else 0}")
                
                if selected_faces and 0 <= face_index < len(selected_faces):
                    selected_face = selected_faces[face_index]
                #   print(f"[DEBUG] Swapping face at index {face_index} (mode: one, order: {face_selector_order})")
                    result_frame = target_vision_frame.copy()
                    result_frame = swap_face(source_face, selected_face, result_frame)
                    print(f"[DEBUG] Swap successful, result shape: {result_frame.shape}")
                    return result_frame
                else:
                    print(f"[DEBUG] No valid face at index {face_index} — no swap")
                    # Вместо logger.info: записать в файл
                    error_message = wording.get('no_valid_face') % face_index
                    error_file = os.path.join(tempfile.gettempdir(), 'facefusion_error.txt')
                    with open(error_file, 'w') as f:
                        f.write(error_message)
            except Exception as e:
            #   print(f"[DEBUG] Error in 'one' mode: {str(e)}")
                import traceback
            #   print(f"[DEBUG] Traceback: {traceback.format_exc()}")
        else:
            print(f"[DEBUG] Unknown mode '{face_selector_mode}' — no swap")
        
    #   print("[DEBUG] No swap performed, returning original frame")
        return target_vision_frame
    
    except Exception as e:
    #   print(f"[ERROR] process_frame crashed: {str(e)}")
        import traceback
    #   print(f"[ERROR] Traceback: {traceback.format_exc()}")
        return target_vision_frame if 'target_vision_frame' in locals() else None


def process_frames(source_paths : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
	source_frames = read_static_images(source_paths)
	source_faces = []

	for source_frame in source_frames:
		temp_faces = get_many_faces([ source_frame ])
		temp_faces = sort_faces_by_order(temp_faces, 'large-small')
		if temp_faces:
			source_faces.append(get_first(temp_faces))
	source_face = get_average_face(source_faces)

	for queue_payload in process_manager.manage(queue_payloads):
		target_vision_path = queue_payload['frame_path']
		target_vision_frame = read_image(target_vision_path)
		output_vision_frame = process_frame(
		{
			'reference_faces': reference_faces,
			'source_face': source_face,
			'target_vision_frame': target_vision_frame
		})
		write_image(target_vision_path, output_vision_frame)
		update_progress(1)


def process_image(source_paths: List[str], target_path: str, output_path: str) -> None:
#   print(f"[DEBUG] Processing image with source paths: {source_paths}")
    reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
    source_frames = read_static_images(source_paths)
#   print(f"[DEBUG] Read {len(source_frames)} source frames")
    source_faces = []

    for source_frame in source_frames:
    #   print(f"[DEBUG] Processing source frame, shape: {source_frame.shape if source_frame is not None else 'None'}")
        temp_faces = get_many_faces([source_frame])
    #   print(f"[DEBUG] Found {len(temp_faces)} faces in source frame")
        temp_faces = sort_faces_by_order(temp_faces, 'large-small')
        if temp_faces:
            source_faces.append(get_first(temp_faces))
    
    source_face = get_average_face(source_faces)
#   print(f"[DEBUG] Got average face: {source_face is not None}")
    
    target_vision_frame = read_static_image(target_path)
#   print(f"[DEBUG] Read target frame, shape: {target_vision_frame.shape if target_vision_frame is not None else 'None'}")
    
    output_vision_frame = process_frame({
        'reference_faces': reference_faces,
        'source_face': source_face,
        'target_vision_frame': target_vision_frame
    })
    write_image(output_path, output_vision_frame)


def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
	processors.multi_process_frames(source_paths, temp_frame_paths, process_frames)