Function generate

app.py:87–188

(
    image_input: str, 
    pose_input: str, 
    width: int, 
    height: int, 
    guidance_scale: float, 
    num_inference_steps: int, 
    fps: int, 
    frames_overlap: int, 
    tile_size: int, 
    noise_aug_strength: float, 
    decode_chunk_size: int,
    seed: int,
)

def generate(
    image_input: str, 
    pose_input: str, 
    width: int, 
    height: int, 
    guidance_scale: float, 
    num_inference_steps: int, 
    fps: int, 
    frames_overlap: int, 
    tile_size: int, 
    noise_aug_strength: float, 
    decode_chunk_size: int,
    seed: int,
):
    gc.collect()
    torch.cuda.empty_cache()
    torch.cuda.ipc_collect()
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    output_dir = Path("outputs")
    output_dir = os.path.join(output_dir, timestamp)
    if seed == -1:
        seed = random.randint(1, 2**20 - 1)
    generator = torch.Generator(device=device).manual_seed(seed)

    pipeline = InferenceAnimationPipeline(
        vae=vae,
        image_encoder=image_encoder,
        unet=unet,
        scheduler=noise_scheduler,
        feature_extractor=feature_extractor,
        pose_net=pose_net,
        face_encoder=face_encoder,
    ).to(device=device, dtype=dtype)

    validation_image_path = image_input
    validation_image = Image.open(image_input).convert('RGB')
    validation_control_images = load_images_from_folder(pose_input, width=width, height=height)

    num_frames = len(validation_control_images)
    face_model.face_helper.clean_all()
    validation_face = cv2.imread(validation_image_path)
    validation_image_bgr = cv2.cvtColor(validation_face, cv2.COLOR_RGB2BGR)
    validation_image_face_info = face_model.app.get(validation_image_bgr)
    if len(validation_image_face_info) > 0:
        validation_image_face_info = sorted(validation_image_face_info, key=lambda x: (x['bbox'][2] - x['bbox'][0]) * (x['bbox'][3] - x['bbox'][1]))[-1]
        validation_image_id_ante_embedding = validation_image_face_info['embedding']
    else:
        validation_image_id_ante_embedding = None

    if validation_image_id_ante_embedding is None:
        face_model.face_helper.read_image(validation_image_bgr)
        face_model.face_helper.get_face_landmarks_5(only_center_face=True)
        face_model.face_helper.align_warp_face()

        if len(face_model.face_helper.cropped_faces) == 0:
            validation_image_id_ante_embedding = np.zeros((512,))
        else:
            validation_image_align_face = face_model.face_helper.cropped_faces[0]