TAESD fix

    elif approximation == 1:

        x_sample = sd_vae_approx.model()(sample.to(devices.device, devices.dtype).unsqueeze(0))[0].detach()

    elif approximation == 3:

        x_sample = sd_vae_taesd.model()(sample.to(devices.device, devices.dtype).unsqueeze(0))[0].detach()

        x_sample = sd_vae_taesd.TAESD.unscale_latents(x_sample)  # returns value in [-2, 2]

        x_sample = x_sample * 0.5

        x_sample = sample * 1.5

        x_sample = sd_vae_taesd.model()(x_sample.to(devices.device, devices.dtype).unsqueeze(0))[0].detach()

    else:

        x_sample = processing.decode_first_stage(shared.sd_model, sample.unsqueeze(0))[0]

    x_sample = torch.clamp((x_sample + 1.0) / 2.0, min=0.0, max=1.0)

    if approximation != 3:

        x_sample = (x_sample + 1.0) / 2.0

    x_sample = torch.clamp(x_sample, min=0.0, max=1.0)

    x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)

    x_sample = x_sample.astype(np.uint8)

class TAESD(nn.Module):

    latent_magnitude = 2

    latent_magnitude = 3

    latent_shift = 0.5

    def __init__(self, decoder_path="taesd_decoder.pth"):