AI Image and Audio Generation Improvements.
AI Video Generation Pre-Flight Checklist. Cost Estimate Improvements.
This commit is contained in:
ajaysi
@@ -88,14 +88,49 @@ class YouTubeVideoRendererService:
|
||||
# Clamp duration to valid WAN 2.5 values (5 or 10 seconds)
|
||||
duration = 5 if duration_estimate <= 7 else 10
|
||||
|
||||
# Log asset usage status
|
||||
has_existing_image = bool(scene.get("imageUrl"))
|
||||
has_existing_audio = bool(scene.get("audioUrl"))
|
||||
|
||||
logger.info(
|
||||
f"[YouTubeRenderer] Rendering scene {scene_number}: "
|
||||
f"resolution={resolution}, duration={duration}s, prompt_length={len(visual_prompt)}"
|
||||
f"resolution={resolution}, duration={duration}s, prompt_length={len(visual_prompt)}, "
|
||||
f"has_existing_image={has_existing_image}, has_existing_audio={has_existing_audio}"
|
||||
)
|
||||
|
||||
# Generate audio if requested - only if narration is not empty
|
||||
# Use existing audio if available, otherwise generate if requested
|
||||
audio_base64 = None
|
||||
if generate_audio_enabled and narration and len(narration.strip()) > 0:
|
||||
scene_audio_url = scene.get("audioUrl")
|
||||
|
||||
if scene_audio_url:
|
||||
# Load existing audio from URL
|
||||
try:
|
||||
from pathlib import Path
|
||||
from urllib.parse import urlparse
|
||||
|
||||
# Extract filename from URL (e.g., /api/youtube/audio/filename.mp3)
|
||||
parsed_url = urlparse(scene_audio_url)
|
||||
audio_filename = Path(parsed_url.path).name
|
||||
|
||||
# Load audio file
|
||||
base_dir = Path(__file__).parent.parent.parent.parent
|
||||
youtube_audio_dir = base_dir / "youtube_audio"
|
||||
audio_path = youtube_audio_dir / audio_filename
|
||||
|
||||
if audio_path.exists():
|
||||
with open(audio_path, "rb") as f:
|
||||
audio_bytes = f.read()
|
||||
audio_base64 = base64.b64encode(audio_bytes).decode('utf-8')
|
||||
logger.info(f"[YouTubeRenderer] Using existing audio for scene {scene_number} from {audio_filename}")
|
||||
else:
|
||||
logger.warning(f"[YouTubeRenderer] Audio file not found: {audio_path}, will generate new audio")
|
||||
raise FileNotFoundError(f"Audio file not found: {audio_path}")
|
||||
except Exception as e:
|
||||
logger.warning(f"[YouTubeRenderer] Failed to load existing audio: {e}, will generate new audio")
|
||||
scene_audio_url = None # Fall back to generation
|
||||
|
||||
# Generate audio if not available and generation is enabled
|
||||
if not audio_base64 and generate_audio_enabled and narration and len(narration.strip()) > 0:
|
||||
try:
|
||||
audio_result = generate_audio(
|
||||
text=narration,
|
||||
@@ -106,7 +141,7 @@ class YouTubeVideoRendererService:
|
||||
audio_bytes = audio_result.audio_bytes if hasattr(audio_result, "audio_bytes") else audio_result
|
||||
# Convert to base64 (just the base64 string, not data URI)
|
||||
audio_base64 = base64.b64encode(audio_bytes).decode('utf-8')
|
||||
logger.info(f"[YouTubeRenderer] Generated audio for scene {scene_number}")
|
||||
logger.info(f"[YouTubeRenderer] Generated new audio for scene {scene_number}")
|
||||
except Exception as e:
|
||||
logger.warning(f"[YouTubeRenderer] Audio generation failed: {e}, continuing without audio")
|
||||
|
||||
@@ -352,6 +387,7 @@ class YouTubeVideoRendererService:
|
||||
self,
|
||||
scenes: List[Dict[str, Any]],
|
||||
resolution: str = "720p",
|
||||
image_model: str = "ideogram-v3-turbo",
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Estimate the cost of rendering a video before actually rendering it.
|
||||
@@ -369,8 +405,16 @@ class YouTubeVideoRendererService:
|
||||
"720p": 0.10,
|
||||
"1080p": 0.15,
|
||||
}
|
||||
|
||||
|
||||
price_per_second = pricing.get(resolution, 0.10)
|
||||
|
||||
# Image generation pricing
|
||||
image_pricing = {
|
||||
"ideogram-v3-turbo": 0.10,
|
||||
"qwen-image": 0.05,
|
||||
}
|
||||
|
||||
image_cost_per_scene = image_pricing.get(image_model, 0.10)
|
||||
|
||||
# Filter enabled scenes
|
||||
enabled_scenes = [s for s in scenes if s.get("enabled", True)]
|
||||
@@ -378,7 +422,8 @@ class YouTubeVideoRendererService:
|
||||
scene_costs = []
|
||||
total_cost = 0.0
|
||||
total_duration = 0.0
|
||||
|
||||
total_image_cost = len(enabled_scenes) * image_cost_per_scene
|
||||
|
||||
for scene in enabled_scenes:
|
||||
scene_number = scene.get("scene_number", 0)
|
||||
duration_estimate = scene.get("duration_estimate", 5)
|
||||
@@ -396,7 +441,10 @@ class YouTubeVideoRendererService:
|
||||
|
||||
total_cost += scene_cost
|
||||
total_duration += duration
|
||||
|
||||
|
||||
# Add image costs to total
|
||||
total_cost += total_image_cost
|
||||
|
||||
return {
|
||||
"resolution": resolution,
|
||||
"price_per_second": price_per_second,
|
||||
@@ -408,5 +456,8 @@ class YouTubeVideoRendererService:
|
||||
"min": round(total_cost * 0.9, 2), # 10% buffer
|
||||
"max": round(total_cost * 1.1, 2), # 10% buffer
|
||||
},
|
||||
"image_model": image_model,
|
||||
"image_cost_per_scene": image_cost_per_scene,
|
||||
"total_image_cost": round(total_image_cost, 2),
|
||||
}
|
||||
|
||||
|
||||
@@ -140,61 +140,87 @@ class YouTubeSceneBuilderService:
|
||||
|
||||
scene_duration_range = duration_metadata.get("scene_duration_range", (5, 15))
|
||||
|
||||
scene_generation_prompt = f"""You are an expert video scriptwriter. Create detailed scenes for a YouTube video based on this plan.
|
||||
scene_generation_prompt = f"""You are a top YouTube scriptwriter specializing in engaging, viral content. Create compelling scenes that captivate viewers and maximize watch time.
|
||||
|
||||
**Video Plan:**
|
||||
- Summary: {video_plan.get('video_summary', '')}
|
||||
- Goal: {video_plan.get('video_goal', '')}
|
||||
- Key Message: {video_plan.get('key_message', '')}
|
||||
- Visual Style: {visual_style}
|
||||
- Tone: {tone}
|
||||
**VIDEO PLAN:**
|
||||
📝 Summary: {video_plan.get('video_summary', '')}
|
||||
🎯 Goal: {video_plan.get('video_goal', '')}
|
||||
💡 Key Message: {video_plan.get('key_message', '')}
|
||||
🎨 Visual Style: {visual_style}
|
||||
🎭 Tone: {tone}
|
||||
|
||||
**Hook Strategy:**
|
||||
**🎣 HOOK STRATEGY:**
|
||||
{hook_strategy}
|
||||
|
||||
**Content Outline:**
|
||||
{chr(10).join([f"- {section.get('section', '')}: {section.get('description', '')} ({section.get('duration_estimate', 0)}s)" for section in content_outline])}
|
||||
**📋 CONTENT STRUCTURE:**
|
||||
{chr(10).join([f"• {section.get('section', '')}: {section.get('description', '')} ({section.get('duration_estimate', 0)}s)" for section in content_outline])}
|
||||
|
||||
**Call-to-Action:**
|
||||
**🚀 CALL-TO-ACTION:**
|
||||
{call_to_action}
|
||||
|
||||
**Duration Constraints:**
|
||||
- Scene duration: {scene_duration_range[0]}-{scene_duration_range[1]} seconds each
|
||||
- Total target: {duration_metadata.get('target_seconds', 150)} seconds
|
||||
**⏱️ TIMING CONSTRAINTS:**
|
||||
• Scene duration: {scene_duration_range[0]}-{scene_duration_range[1]} seconds each
|
||||
• Total target: {duration_metadata.get('target_seconds', 150)} seconds
|
||||
|
||||
**Your Task:**
|
||||
Create detailed scenes that include:
|
||||
1. Scene number and title
|
||||
2. Narration text (what will be spoken)
|
||||
3. Visual description (what viewers will see)
|
||||
4. Duration estimate
|
||||
5. Emphasis tags (hook, main_content, transition, cta)
|
||||
**🎬 YOUR MISSION - CREATE VIRAL-WORTHY SCENES:**
|
||||
|
||||
**Format as JSON array:**
|
||||
Write narration that:
|
||||
✨ **HOOKS IMMEDIATELY** - First {duration_metadata.get('hook_seconds', 10)}s must GRAB attention
|
||||
🎭 **TELLS A STORY** - Each scene advances the narrative with emotional engagement
|
||||
💡 **DELIVERS VALUE** - Provide insights, tips, or "aha!" moments in every scene
|
||||
🔥 **BUILDS EXCITEMENT** - Use power words, questions, and cliffhangers
|
||||
👥 **CONNECTS PERSONALLY** - Speak directly to the viewer's needs and desires
|
||||
⚡ **MAINTAINS PACE** - Vary sentence length for natural rhythm
|
||||
🎯 **DRIVES ACTION** - Build toward the CTA with increasing urgency
|
||||
|
||||
**REQUIRED SCENE ELEMENTS:**
|
||||
1. **scene_number**: Sequential numbering
|
||||
2. **title**: Catchy, descriptive title (5-8 words max)
|
||||
3. **narration**: ENGAGING spoken script with:
|
||||
- Conversational language ("you know what I mean?")
|
||||
- Rhetorical questions ("Have you ever wondered...?")
|
||||
- Power transitions ("But here's the game-changer...")
|
||||
- Emotional hooks ("Imagine this...")
|
||||
- Action-oriented language ("Let's dive in...")
|
||||
4. **visual_description**: Cinematic, professional YouTube visuals
|
||||
5. **duration_estimate**: Realistic speaking time
|
||||
6. **emphasis**: hook/main_content/transition/cta
|
||||
7. **visual_cues**: ["dramatic_zoom", "text_overlay", "fast_cuts"]
|
||||
|
||||
**🎯 YOUTUBE OPTIMIZATION RULES:**
|
||||
• **Hook Power**: First 3 seconds = make them stay or lose them
|
||||
• **Value Density**: Every 10 seconds must deliver new insight
|
||||
• **Emotional Arc**: Build curiosity → teach → inspire → convert
|
||||
• **Natural Flow**: Scenes must connect seamlessly
|
||||
• **CTA Momentum**: Final scene creates irresistible urge to act
|
||||
|
||||
**📊 FORMAT AS JSON ARRAY:**
|
||||
[
|
||||
{{
|
||||
"scene_number": 1,
|
||||
"title": "Hook - Attention Grabber",
|
||||
"narration": "The spoken text for this scene...",
|
||||
"visual_description": "Detailed description of what viewers see...",
|
||||
"duration_estimate": 5,
|
||||
"title": "The Shocking Truth They Hide",
|
||||
"narration": "You won't believe what just happened in my latest discovery! I was scrolling through the usual content when BAM - this completely changed everything I thought about [topic]. And get this - it could transform YOUR results too!",
|
||||
"visual_description": "Dynamic opening shot with shocking text overlay, fast cuts of social media feeds, energetic music swell, close-up of surprised reaction",
|
||||
"duration_estimate": 8,
|
||||
"emphasis": "hook",
|
||||
"visual_cues": ["close-up", "dynamic", "bright"]
|
||||
"visual_cues": ["shocking_text", "fast_cuts", "music_swell", "reaction_shot"]
|
||||
}},
|
||||
...
|
||||
]
|
||||
|
||||
Make sure:
|
||||
- First scene is a strong hook ({duration_metadata.get('hook_seconds', 10)}s)
|
||||
- Last scene includes the CTA ({duration_metadata.get('cta_seconds', 10)}s)
|
||||
- Each scene has clear narration and visual description
|
||||
- Total duration fits within {duration_metadata.get('target_seconds', 150)} seconds
|
||||
- Scenes flow naturally from one to the next
|
||||
"""
|
||||
**🔥 SUCCESS CRITERIA:**
|
||||
✅ First scene hooks in 3 seconds
|
||||
✅ Each scene delivers 1-2 key insights
|
||||
✅ Narration feels like talking to a friend
|
||||
✅ Total story arc creates emotional journey
|
||||
✅ CTA feels like the natural next step
|
||||
✅ Scenes fit duration perfectly"""
|
||||
|
||||
system_prompt = (
|
||||
"You are an expert video scriptwriter specializing in YouTube content. "
|
||||
"Your scenes are engaging, well-paced, and optimized for viewer retention."
|
||||
"You are a master YouTube scriptwriter who creates viral, engaging content that "
|
||||
"keeps viewers watching until the end. You understand YouTube algorithm optimization, "
|
||||
"emotional storytelling, and creating irresistible hooks that make viewers hit 'like' and 'subscribe'. "
|
||||
"Your scripts are conversational, valuable, and conversion-focused."
|
||||
)
|
||||
|
||||
response = llm_text_gen(
|
||||
|
||||
Reference in New Issue
Block a user