Update ltx_worker_upscaler.py

ltx_worker_upscaler.py

@@ -12,10 +12,11 @@ import numpy as np
 import imageio
 from pathlib import Path
 import huggingface_hub
+from PIL import Image # <--- IMPORTAÇÃO ADICIONADA AQUI
 
 from inference import create_ltx_video_pipeline
 from ltx_video.models.autoencoders.latent_upsampler import LatentUpsampler
-from ltx_video.models.autoencoders.vae_encode import vae_decode
+from ltx_video.models.autoencoders.vae_encode import vae_encode, vae_decode
 
 class LtxUpscaler:
     def __init__(self, device_id='cuda:0'):
@@ -75,15 +76,10 @@ class LtxUpscaler:
     def upscale_latents_to_video(self, latent_path: str, output_path: str, video_fps: int):
         print(f"UPSCALER ({self.device}): Processando latentes de {os.path.basename(latent_path)}")
         
-        # Carrega os latentes do disco e os envia para a GPU
         latents = torch.load(latent_path).to(self.device, dtype=self.model_dtype)
 
-        # PASSO 1: Upscale Espacial (não precisamos mais de vae_encode)
         upsampled_latents = self.latent_upsampler(latents)
         
-        # (Opcional: PASSO 2 - Upscale Temporal seria inserido aqui no futuro)
-
-        # PASSO 3: Decodificação Final
         decode_timestep = torch.tensor([0.0] * upsampled_latents.shape[0], device=self.device)
         upsampled_video_tensor = vae_decode(
             upsampled_latents, self.vae, is_video=True, timestep=decode_timestep
@@ -110,7 +106,6 @@ class LtxUpscaler:
         )
         
         decoded_tensor = (decoded_tensor.clamp(-1, 1) + 1) / 2.0
-        # Shape: (B, C, F, H, W) -> (H, W, C)
         numpy_image = (decoded_tensor[0].permute(2, 3, 1, 0).squeeze().cpu().float().numpy() * 255).astype(np.uint8)
         return Image.fromarray(numpy_image)
 #--- END OF MODIFIED FILE app_fluxContext_Ltx/ltx_worker_upscaler.py ---