Create llm_inference_video.py

llm_inference_video.py

@@ -0,0 +1,125 @@
+import os
+import random
+from groq import Groq
+from openai import OpenAI
+from gradio_client import Client
+
+class VideoLLMInferenceNode:
+    def __init__(self):
+        self.huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
+        self.groq_api_key = os.getenv("GROQ_API_KEY")
+        self.sambanova_api_key = os.getenv("SAMBANOVA_API_KEY")
+        
+        self.huggingface_client = OpenAI(
+            base_url="https://api-inference.huggingface.co/v1/",
+            api_key=self.huggingface_token,
+        )
+        self.groq_client = Groq(api_key=self.groq_api_key)
+        self.sambanova_client = OpenAI(
+            api_key=self.sambanova_api_key,
+            base_url="https://api.sambanova.ai/v1",
+        )
+
+    def generate_video_prompt(
+        self,
+        input_concept,
+        duration,
+        style,
+        camera_style,
+        pacing,
+        special_effects,
+        custom_elements,
+        provider="Hugging Face",
+        model=None
+    ):
+        try:
+            # Video prompt templates
+            prompt_templates = {
+                "cinematic": f"""Create a detailed cinematic prompt for a {duration}-second video. Include:
+- 3-5 distinct scenes with smooth transitions
+- Camera movements: {camera_style}
+- Lighting design for {style} style
+- Special effects: {special_effects}
+- Color grading and film grain details
+- Pacing: {pacing}
+- Add {custom_elements if custom_elements else 'unique atmospheric elements'}
+Format: Timestamped scene descriptions with shot types and transition notes.""",
+
+                "documentary": f"""Develop a documentary-style video prompt for {duration} seconds. Include:
+- Interview setup with lighting and background
+- B-roll sequences (3-5 locations)
+- Archival footage integration
+- Text overlay and info-graphics
+- Narration style and tone
+- {camera_style} camera work
+- {pacing} rhythm for topic exploration
+- {special_effects} for historical recreations""",
+
+                "animation": f"""Create a {style} animation prompt for {duration} seconds. Specify:
+- Animation technique (2D/3D/stop-motion)
+- Key action sequences (3-5)
+- Character design elements
+- Background art style
+- Motion blur and frame rate considerations
+- Camera zooms/pans for {pacing} pacing
+- Special effects: {special_effects}
+- {custom_elements if custom_elements else 'unique stylistic flourishes'}""",
+
+                "action": f"""Generate intense action sequence prompt ({duration} seconds). Include:
+- 3 escalating action beats
+- Camera angles for {style} impact
+- Stunt choreography details
+- Slow-motion/fast-cut ratios
+- Explosion/sfx elements: {special_effects}
+- Pacing structure: {pacing}
+- {camera_style} camera movements
+- Hero shot composition""",
+
+                "experimental": f"""Design avant-garde video prompt ({duration} seconds) with:
+- Unconventional narrative structure
+- {style} visual treatments
+- Abstract transitions between {random.randint(5,8)} concepts
+- Experimental sound/image relationships
+- {camera_style} capture techniques
+- {special_effects} digital manipulations
+- Pacing: {pacing} with {custom_elements if custom_elements else 'temporal distortions'}"""
+            }
+
+            base_prompt = prompt_templates.get(style.lower(), prompt_templates["cinematic"])
+            system_message = """You are a professional video director and cinematography expert. 
+Generate rich, technical video prompts that include:
+1. Scene-by-scene breakdowns with timestamps
+2. Camera movements and lens specifications
+3. Lighting setups and color palettes
+4. Transition types and durations
+5. Special effects implementation
+6. Pacing and rhythm markers
+7. Technical specifications when relevant"""
+
+            # Select provider
+            if provider == "Hugging Face":
+                client = self.huggingface_client
+                model = model or "meta-llama/Meta-Llama-3.1-70B-Instruct"
+            elif provider == "Groq":
+                client = self.groq_client
+                model = model or "llama-3.1-70b-versatile"
+            elif provider == "SambaNova":
+                client = self.sambanova_client
+                model = model or "Meta-Llama-3.1-70B-Instruct"
+
+            response = client.chat.completions.create(
+                model=model,
+                messages=[
+                    {"role": "system", "content": system_message},
+                    {"role": "user", "content": f"{base_prompt}\nCore Concept: {input_concept}"}
+                ],
+                temperature=1.2,
+                max_tokens=1500,
+                top_p=0.95,
+                seed=random.randint(0, 10000)
+            )
+
+            return response.choices[0].message.content.strip()
+
+        except Exception as e:
+            return f"Error generating video prompt: {str(e)}"