Update api/ltx/ltx_aduc_orchestrator.py

api/ltx/ltx_aduc_orchestrator.py

@@ -1,20 +1,18 @@
 # FILE: api/ltx_aduc_orchestrator.py
 # DESCRIPTION: The main workflow orchestrator for the ADUC-SDR LTX suite.
-# It acts as the primary entry point for the UI, coordinating the specialized
-# LTX and VAE clients to execute a complete video generation pipeline from prompt to MP4.
+# In this simplified architecture, it coordinates a single unified client (LtxAducPipeline)
+# to execute the complete video generation pipeline from prompt to MP4.
 
 import logging
 import time
 import yaml
 from PIL import Image
 from typing import Optional, Dict
-import os
 
-# O Orquestrador importa os CLIENTES especialistas que ele vai coordenar.
+# O Orquestrador importa o CLIENTE UNIFICADO que ele vai coordenar.
 from api.ltx.ltx_aduc_pipeline import ltx_aduc_pipeline
-from api.ltx.vae_aduc_pipeline import vae_aduc_pipeline
 
-# O Orquestrador importa as FERRAMENTAS de que precisa.
+# O Orquestrador importa as FERRAMENTAS de que precisa para as tarefas finais.
 from tools.video_encode_tool import video_encode_tool_singleton
 
 # Importa o Path para carregar a configuração.
@@ -27,8 +25,8 @@ LTX_VIDEO_REPO_DIR = Path("/data/LTX-Video")
 
 class LtxAducOrchestrator:
     """
-    Orquestra o fluxo de trabalho completo de geração de vídeo.
-    É o ponto de entrada principal para a UI, preparando e delegando tarefas.
+    Orquestra o fluxo de trabalho completo de geração de vídeo,
+    coordenando o cliente unificado LTX. É o ponto de entrada principal para a UI.
     """
     def __init__(self):
         """
@@ -86,16 +84,14 @@ class LtxAducOrchestrator:
 
             initial_conditioning_items = []
             if initial_image:
-                logging.info("Delegating to VAE client: create conditioning item.")
+                logging.info("Orchestrator delegating: create conditioning item.")
                 conditioning_params = [(0, 1.0)] # (frame_number, strength)
-                initial_conditioning_items = vae_aduc_pipeline(
-                    media=[initial_image],
-                    task='create_conditioning_items',
-                    target_resolution=(height, width),
-                    conditioning_params=conditioning_params
+                initial_conditioning_items = ltx_aduc_pipeline.encode_to_conditioning_items(
+                    media_list=[initial_image],
+                    params=conditioning_params,
+                    resolution=(height, width)
                 )
 
-            # Prepara a lista de argumentos comuns para todos os chunks de LTX.
             common_ltx_args = self.base_config.get("first_pass", {}).copy()
             common_ltx_args.update({
                 'negative_prompt': "blurry, low quality, bad anatomy, deformed",
@@ -108,8 +104,8 @@ class LtxAducOrchestrator:
             # =================================================================
             # --- ETAPA 2: DELEGAR GERAÇÃO DO VÍDEO LATENTE ---
             # =================================================================
-            logging.info("Delegating to LTX client: generate latent video.")
-            final_latents, used_seed = ltx_aduc_pipeline(
+            logging.info("Orchestrator delegating: generate latent video.")
+            final_latents, used_seed = ltx_aduc_pipeline.generate_latents(
                 prompt_list=prompt_list,
                 duration_in_seconds=duration_in_seconds,
                 common_ltx_args=common_ltx_args,
@@ -117,16 +113,16 @@ class LtxAducOrchestrator:
             )
             if final_latents is None:
                 raise RuntimeError("LTX client failed to generate a latent tensor.")
-            logging.info(f"LTX client returned latent tensor with shape: {final_latents.shape}")
+            logging.info(f"Orchestrator received latent tensor with shape: {final_latents.shape}")
 
             # =================================================================
             # --- ETAPA 3: DELEGAR DECODIFICAÇÃO PARA PIXELS ---
             # =================================================================
-            logging.info("Delegating to VAE client: decode latent to pixels.")
-            pixel_tensor = vae_aduc_pipeline(media=final_latents, task='decode')
+            logging.info("Orchestrator delegating: decode latent to pixels.")
+            pixel_tensor = ltx_aduc_pipeline.decode_to_pixels(final_latents)
             if pixel_tensor is None:
-                raise RuntimeError("VAE client failed to decode the latent tensor.")
-            logging.info(f"VAE client returned pixel tensor with shape: {pixel_tensor.shape}")
+                raise RuntimeError("LTX client failed to decode the latent tensor.")
+            logging.info(f"Orchestrator received pixel tensor with shape: {pixel_tensor.shape}")
 
             # =================================================================
             # --- ETAPA 4: TAREFA FINAL - CODIFICAR PARA MP4 ---
@@ -134,7 +130,7 @@ class LtxAducOrchestrator:
             video_filename = f"{output_filename_base}_{int(time.time())}_{used_seed}.mp4"
             output_path = f"{self.output_dir}/{video_filename}"
             
-            logging.info(f"Executing final task: saving tensor to MP4 at {output_path}")
+            logging.info(f"Orchestrator executing final task: saving tensor to MP4 at {output_path}")
             video_encode_tool_singleton.save_video_from_tensor(
                 pixel_5d=pixel_tensor,
                 path=output_path,
@@ -152,10 +148,9 @@ class LtxAducOrchestrator:
 
 # ==============================================================================
 # --- INSTÂNCIA SINGLETON DO ORQUESTRADOR ---
-# Este é o ponto de entrada principal que a UI (app.py) irá chamar.
 # ==============================================================================
 try:
     ltx_aduc_orchestrator = LtxAducOrchestrator()
-except Exception as e:
+except Exception:
     logging.critical("CRITICAL: Failed to initialize the LtxAducOrchestrator.", exc_info=True)
     ltx_aduc_orchestrator = None