File size: 15,060 Bytes
b12e499 404465b b12e499 869d082 b12e499 869d082 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 |
"""
LLM-powered script generation for EceMotion Pictures.
Generates intelligent, structure-aware commercial scripts with timing markers.
"""
import logging
import random
from typing import Dict, List, Optional, Tuple
from dataclasses import dataclass
from config import (
MODEL_LLM, MODEL_CONFIGS, VOICE_STYLES, STRUCTURE_TEMPLATES, TAGLINES,
get_safe_model_name
)
logger = logging.getLogger(__name__)
@dataclass
class ScriptSegment:
"""Represents a segment of the commercial script with timing information."""
text: str
duration_estimate: float
segment_type: str # "hook", "flow", "benefit", "cta"
timing_marker: Optional[str] = None
@dataclass
class GeneratedScript:
"""Complete generated script with all segments and metadata."""
segments: List[ScriptSegment]
total_duration: float
tagline: str
voice_style: str
word_count: int
raw_script: str
class LLMScriptGenerator:
"""Generates commercial scripts using large language models with fallbacks."""
def __init__(self, model_name: str = MODEL_LLM):
self.model_name = get_safe_model_name(model_name, "llm")
self.model = None
self.tokenizer = None
self.model_config = MODEL_CONFIGS.get(self.model_name, {})
self.llm_available = False
# Try to initialize LLM
self._try_init_llm()
def _try_init_llm(self):
"""Try to initialize the LLM model."""
try:
if "dialo" in self.model_name.lower():
self._init_dialogpt()
elif "qwen" in self.model_name.lower():
self._init_qwen()
else:
logger.warning(f"Unknown LLM model: {self.model_name}, using fallback")
self.llm_available = False
except Exception as e:
logger.warning(f"Failed to initialize LLM {self.model_name}: {e}")
self.llm_available = False
def _init_dialogpt(self):
"""Initialize DialoGPT model."""
try:
from transformers import AutoTokenizer, AutoModelForCausalLM
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
if self.tokenizer.pad_token is None:
self.tokenizer.pad_token = self.tokenizer.eos_token
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
torch_dtype="auto",
device_map="auto" if self._has_gpu() else "cpu"
)
self.llm_available = True
logger.info(f"DialoGPT model {self.model_name} loaded successfully")
except Exception as e:
logger.error(f"Failed to load DialoGPT: {e}")
self.llm_available = False
def _init_qwen(self):
"""Initialize Qwen model."""
try:
from transformers import AutoTokenizer, AutoModelForCausalLM
self.tokenizer = AutoTokenizer.from_pretrained(
self.model_name,
trust_remote_code=True
)
if self.tokenizer.pad_token is None:
self.tokenizer.pad_token = self.tokenizer.eos_token
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
torch_dtype="auto",
device_map="auto" if self._has_gpu() else "cpu",
trust_remote_code=True
)
self.llm_available = True
logger.info(f"Qwen model {self.model_name} loaded successfully")
except Exception as e:
logger.error(f"Failed to load Qwen: {e}")
self.llm_available = False
def _has_gpu(self) -> bool:
"""Check if GPU is available."""
try:
import torch
return torch.cuda.is_available()
except ImportError:
return False
def _create_system_prompt(self) -> str:
"""Create system prompt for retro commercial script generation."""
return """You are a professional copywriter specializing in 1980s-style TV commercials.
Your task is to create engaging, persuasive commercial scripts that capture the authentic retro aesthetic.
Key requirements:
- Use 1980s commercial language and style
- Include clear hooks, benefits, and calls-to-action
- Keep scripts concise and punchy
- Use active voice and emotional appeals
- End with a memorable tagline
Format your response as:
HOOK: [Opening attention-grabber]
FLOW: [Main content following the structure]
BENEFIT: [Key value proposition]
CTA: [Call to action with tagline]
Keep each segment under 2-3 sentences. Use enthusiastic, confident language typical of 1980s advertising."""
def _create_user_prompt(self, brand: str, structure: str, script_prompt: str,
duration: int, voice_style: str) -> str:
"""Create user prompt with specific requirements."""
return f"""Create a {duration}-second retro commercial script for {brand}.
Structure: {structure}
Script idea: {script_prompt}
Voice style: {voice_style}
Make it authentic to 1980s TV commercials with the energy and style of that era."""
def _parse_script_response(self, response: str) -> List[ScriptSegment]:
"""Parse LLM response into structured script segments."""
segments = []
# Split by segment markers
import re
parts = re.split(r'(HOOK:|FLOW:|BENEFIT:|CTA:)', response)
for i in range(1, len(parts), 2):
if i + 1 < len(parts):
segment_type = parts[i].rstrip(':').lower()
text = parts[i + 1].strip()
if text:
# Estimate duration based on word count (150 WPM)
word_count = len(text.split())
duration = (word_count / 150) * 60 # Convert to seconds
segments.append(ScriptSegment(
text=text,
duration_estimate=duration,
segment_type=segment_type,
timing_marker=f"[{segment_type.upper()}]"
))
return segments
def _extract_tagline(self, response: str) -> str:
"""Extract tagline from the script response."""
# Look for tagline in CTA section
import re
cta_match = re.search(r'CTA:.*?([A-Z][^.!?]*[.!?])', response, re.DOTALL)
if cta_match:
cta_text = cta_match.group(1)
# Extract the last sentence as potential tagline
sentences = re.split(r'[.!?]+', cta_text)
if sentences:
tagline = sentences[-1].strip()
if len(tagline) > 5: # Ensure it's substantial
return tagline
# Fallback to predefined taglines
return random.choice(TAGLINES)
def generate_script_with_llm(self, brand: str, structure: str, script_prompt: str,
duration: int, voice_style: str, seed: int = 42) -> GeneratedScript:
"""Generate script using LLM."""
if not self.llm_available:
raise RuntimeError("LLM not available")
# Set random seed for reproducibility
random.seed(seed)
# Create prompts
system_prompt = self._create_system_prompt()
user_prompt = self._create_user_prompt(brand, structure, script_prompt, duration, voice_style)
# Format for the model
if "dialo" in self.model_name.lower():
# DialoGPT format
text = f"{user_prompt}\n\nResponse:"
else:
# Generic format
text = f"System: {system_prompt}\n\nUser: {user_prompt}\n\nAssistant:"
# Tokenize
inputs = self.tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
# Move inputs to same device as model
device = next(self.model.parameters()).device
inputs = {k: v.to(device) for k, v in inputs.items()}
# Generate
self.model.eval()
outputs = self.model.generate(
**inputs,
max_new_tokens=self.model_config.get("max_tokens", 256),
temperature=self.model_config.get("temperature", 0.7),
top_p=self.model_config.get("top_p", 0.9),
do_sample=True,
pad_token_id=self.tokenizer.eos_token_id,
eos_token_id=self.tokenizer.eos_token_id,
num_return_sequences=1
)
# Decode response
response = self.tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
logger.info(f"Generated script response: {response[:200]}...")
# Parse response
segments = self._parse_script_response(response)
tagline = self._extract_tagline(response)
# Calculate total duration
total_duration = sum(segment.duration_estimate for segment in segments)
# Calculate word count
word_count = sum(len(segment.text.split()) for segment in segments)
return GeneratedScript(
segments=segments,
total_duration=total_duration,
tagline=tagline,
voice_style=voice_style,
word_count=word_count,
raw_script=response
)
def generate_script_with_template(self, brand: str, structure: str, script_prompt: str,
duration: int, voice_style: str, seed: int = 42) -> GeneratedScript:
"""Generate script using template-based approach (fallback)."""
random.seed(seed)
# Select structure template
structure_template = structure.strip() or random.choice(STRUCTURE_TEMPLATES)
# Generate segments based on template
segments = []
# Hook
hook_text = script_prompt or f"Introducing {brand} - the future is here!"
segments.append(ScriptSegment(
text=hook_text,
duration_estimate=2.0,
segment_type="hook",
timing_marker="[HOOK]"
))
# Flow (based on structure)
flow_text = f"With {structure_template.lower()}, {brand} delivers results like never before."
segments.append(ScriptSegment(
text=flow_text,
duration_estimate=3.0,
segment_type="flow",
timing_marker="[FLOW]"
))
# Benefit
benefit_text = "Faster, simpler, cooler - just like your favorite retro tech."
segments.append(ScriptSegment(
text=benefit_text,
duration_estimate=2.5,
segment_type="benefit",
timing_marker="[BENEFIT]"
))
# CTA
tagline = random.choice(TAGLINES)
cta_text = f"Try {brand} today. {tagline}"
segments.append(ScriptSegment(
text=cta_text,
duration_estimate=2.5,
segment_type="cta",
timing_marker="[CTA]"
))
# Calculate totals
total_duration = sum(segment.duration_estimate for segment in segments)
word_count = sum(len(segment.text.split()) for segment in segments)
return GeneratedScript(
segments=segments,
total_duration=total_duration,
tagline=tagline,
voice_style=voice_style,
word_count=word_count,
raw_script=f"Template-based script for {brand}"
)
def generate_script(self, brand: str, structure: str, script_prompt: str,
duration: int, voice_style: str, seed: int = 42) -> GeneratedScript:
"""
Generate a complete commercial script.
"""
try:
if self.llm_available:
return self.generate_script_with_llm(brand, structure, script_prompt, duration, voice_style, seed)
else:
logger.info("Using template-based script generation (LLM not available)")
return self.generate_script_with_template(brand, structure, script_prompt, duration, voice_style, seed)
except Exception as e:
logger.error(f"Script generation failed: {e}")
logger.info("Falling back to template-based generation")
return self.generate_script_with_template(brand, structure, script_prompt, duration, voice_style, seed)
def suggest_scripts(self, structure: str, n: int = 6, seed: int = 0) -> List[str]:
"""
Generate multiple script suggestions based on structure.
"""
try:
suggestions = []
for i in range(n):
script = self.generate_script(
brand="YourBrand",
structure=structure,
script_prompt="Create an engaging hook",
duration=10,
voice_style="Announcer '80s",
seed=seed + i
)
# Extract hook from first segment
if script.segments:
hook = script.segments[0].text
suggestions.append(hook)
else:
suggestions.append("Back to '87 - the future is now!")
return suggestions
except Exception as e:
logger.warning(f"Script suggestion failed: {e}")
# Fallback to original random generation
return self._fallback_suggestions(structure, n, seed)
def _fallback_suggestions(self, structure: str, n: int, seed: int) -> List[str]:
"""Fallback to original random script generation."""
random.seed(seed)
base = (structure or "").lower().strip()
ideas = []
for _ in range(n):
style = random.choice(["infomercial", "mall ad", "late-night", "newsflash", "arcade bumper"])
shot = random.choice(["neon grid", "CRT scanlines", "vaporwave sunset", "shopping mall", "boombox close-up"])
hook = random.choice([
"Remember this sound?", "Back to '87.", "Deal of the decade.",
"We paused time.", "Be kind, rewind your brand."
])
idea = f"{hook} {style} with {shot}."
# Light correlation with structure
for kw in ["montage", "testimonial", "news", "unboxing", "before", "after", "countdown", "logo", "cta"]:
if kw in base and kw not in idea:
idea += f" Includes {kw}."
ideas.append(idea)
return ideas
def create_script_generator() -> LLMScriptGenerator:
"""Factory function to create a script generator."""
return LLMScriptGenerator()
|