Update engineers/deformes3D_thinker.py

engineers/deformes3D_thinker.py

@@ -30,82 +30,50 @@
 #
 # Version 2.1.1
 
+
 import logging
-from pathlib import Path
 from PIL import Image
 import gradio as gr
-import torch
-import numpy as np
 
-from managers.ltx_manager import ltx_manager_singleton
-from ltx_video.utils.prompt_enhance_utils import I2V_CINEMATIC_PROMPT
-from ltx_video.pipelines.pipeline_ltx_video import ConditioningItem
+# A única dependência agora é o nosso novo manager dedicado
+from managers.prompt_enhancer_manager import prompt_enhancer_manager_singleton
 
 logger = logging.getLogger(__name__)
 
 class Deformes3DThinker:
     """
-    The tactical specialist that handles cinematic decision-making by leveraging
-    the LTX pipeline's internal prompt enhancement capabilities.
+    The tactical specialist that handles cinematic decision-making by delegating
+    prompt generation to the specialized PromptEnhancerManager.
     """
     
     def __init__(self):
-        # Acessa a pipeline de enhancement exposta publicamente pelo LTX manager.
-        self.ltx_pipeline = ltx_manager_singleton.prompt_enhancement_pipeline
-        if not self.ltx_pipeline:
-            raise RuntimeError("Deformes3DThinker could not access the LTX prompt enhancement pipeline.")
-        
-        logger.info("Deformes3DThinker initialized and linked to LTX pipeline's enhancement models.")
+        if not prompt_enhancer_manager_singleton:
+            raise RuntimeError("Deformes3DThinker requires the PromptEnhancerManager to be initialized.")
+        self.enhancer = prompt_enhancer_manager_singleton
+        logger.info("Deformes3DThinker initialized and linked to PromptEnhancerManager.")
 
     def get_enhanced_motion_prompt(self, global_prompt: str, story_history: str, 
                                    past_keyframe_path: str, present_keyframe_path: str, future_keyframe_path: str,
                                    past_scene_desc: str, present_scene_desc: str, future_scene_desc: str) -> str:
         """
-        Generates a refined, cinematic motion prompt using the LTX pipeline's own
-        image captioning and LLM enhancement models.
+        Gets a refined, cinematic motion prompt by delegating to the PromptEnhancerManager.
+        It uses the present image as visual context and the future scene description as the goal.
         """
         try:
-            image_caption_model = self.ltx_pipeline.prompt_enhancer_image_caption_model
-            image_caption_processor = self.ltx_pipeline.prompt_enhancer_image_caption_processor
-            llm_model = self.ltx_pipeline.prompt_enhancer_llm_model
-            llm_tokenizer = self.ltx_pipeline.prompt_enhancer_llm_tokenizer
-
-            if not all([image_caption_model, image_caption_processor, llm_model, llm_tokenizer]):
-                logger.warning("LTX prompt enhancement models not found. Using fallback.")
-                return f"A cinematic transition from '{present_scene_desc}' to '{future_scene_desc}'."
-
             present_image = Image.open(present_keyframe_path).convert("RGB")
-            images_list = [present_image]
-            
-            # Gerar a caption da imagem de referência
-            inputs = image_caption_processor(
-                ["<DETAILED_CAPTION>"] * len(images_list), images_list, return_tensors="pt"
-            ).to(image_caption_model.device)
-            generated_ids = image_caption_model.generate(**inputs, max_new_tokens=1024, do_sample=False, num_beams=3)
-            image_captions = image_caption_processor.batch_decode(generated_ids, skip_special_tokens=True)
-            
-            # Criar o prompt para o LLM de enhancement
-            user_content = f"user_prompt: {future_scene_desc}\nimage_caption: {image_captions[0]}"
-            messages = [{"role": "system", "content": I2V_CINEMATIC_PROMPT}, {"role": "user", "content": user_content}]
-            
-            text = llm_tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-            model_inputs = llm_tokenizer([text], return_tensors="pt").to(llm_model.device)
-
-            # Chamar o LLM e decodificar a resposta
-            output_ids = llm_model.generate(**model_inputs, max_new_tokens=256)
-            input_ids_len = model_inputs.input_ids.shape[1]
-            decoded_prompts = llm_tokenizer.batch_decode(output_ids[:, input_ids_len:], skip_special_tokens=True)
-            
-            enhanced_prompt = decoded_prompts[0]
             
-            logger.info(f"Deformes3DThinker (LTX) Decision -> Motion Prompt: '{enhanced_prompt}'")
-            return enhanced_prompt.strip()
+            # Delega a tarefa complexa para o especialista
+            enhanced_prompt = self.enhancer.generate_enhanced_prompt(
+                image=present_image,
+                user_prompt=future_scene_desc
+            )
+            return enhanced_prompt
 
         except Exception as e:
-            logger.error(f"The Film Director (Deformes3D Thinker) failed with LTX models: {e}. Using fallback.", exc_info=True)
+            logger.error(f"The Film Director (Deformes3D Thinker) failed: {e}. Using fallback.", exc_info=True)
             fallback_prompt = f"A smooth, continuous cinematic transition from '{present_scene_desc}' to '{future_scene_desc}'."
             logger.info(f"Deformes3DThinker Fallback -> Motion Prompt: '{fallback_prompt}'")
             return fallback_prompt
 
 # --- Singleton Instance ---
-deformes3d_thinker_singleton = Deformes3DThinker()
+deformes3d_thinker_singleton = Deformes3DThinker()