AI Video Generation

Quick Reference

Model	Provider	Duration	Best For
sora-2	OpenAI	4/8/12s	Yüksek kalite, gerçekçi
sora-2-pro	OpenAI	4/8/12s	Native speech, conversational only
kling-3.0-pro	fal.ai	5/10/15s	Sinematik, fizik tabanlı, native audio

Model Selection (Brain Agent)

Brain Agent otomatik model secimi yapar. Manuel secim icin:

from app.video_models import get_model_config, get_available_models, validate_duration

models = get_available_models()  # ["sora-2", "sora-2-pro", "kling-3.0-pro"]
config = get_model_config("kling-3.0-pro")
duration = validate_duration("kling-3.0-pro", 15)  # -> 15

# Model secim rehberi:
# Haber bazlı acil icerik → kling-3.0-pro (hızlı, 2-3 dk)
# Sinematik showcase → sora-2
# Sesli anlatım → sora-2-pro
# Conversational diyalog → sora-2-pro
# Teknik demo → kling-3.0-pro

Video Style System

from app.video_styles import VIDEO_STYLES, get_style_config

# 10 görsel stil:
# cinematic_4k, anime, cartoon_3d, watercolor, 3d_render,
# pixel_art, comic_book, claymation, minimalist, neon_cyberpunk

style = get_style_config("cinematic_4k")
# -> character descriptions, scene settings, voice type mappings

Generation Functions

# Sora (OpenAI)
from app.sora_helper import generate_video_sora
result = await generate_video_sora(prompt, model="sora-2", size="720x1280", duration=12)

# Kling (via fal.ai)
from app.kling_helper import KlingHelper
kling = KlingHelper()
result = await kling.generate_video(
    prompt=prompt,
    duration=10,          # 5/10/15s
    aspect_ratio="9:16",
    generate_audio=True   # Native audio (fal.ai default)
)

Kling via fal.ai Details

• Endpoint: fal-ai/kling-video/v3/pro/text-to-video
• Auth: FAL_API_KEY (shared with nano-banana and lip-sync)
• Model: Kling 3.0 Pro
• Durations: 5s, 10s, 15s
• Features: Native audio, physics-based motion, cinematic quality
• Pay-as-you-go pricing via fal.ai queue API

Instagram Specs

Spec	Value
Aspect	9:16 vertical
Resolution	720x1280
Codec	H.264/AAC
Max Duration	90s

Environment Variables

OPENAI_API_KEY=...          # Sora
FAL_API_KEY=...             # Kling video, infographics, lip-sync

Return Format

# Success
{"success": True, "video_path": "/path/to.mp4", "model_used": "kling-3.0-pro"}

# Error with fallback
{"success": False, "error": "...", "fallback": "sora"}

Best Practices

• Simple, comma-separated sentences
• Include camera movement + lighting
• Avoid complex physics (bouncing, running)
• Use: "cinematic", "professional", "documentary"
• Duration: 5-8s ideal for Reels
• Kling 3.0 Pro en hizli (2-3 dk), 15s destegi

Production Pipelines

Pipeline	Dosya	Default Model
reels	app/production/reels_pipeline.py	kling-3.0-pro
voice_reels	app/production/voice_reels_pipeline.py	sora-2-pro
news_reels	app/production/news_reels_pipeline.py	kling-3.0-pro
long_video	app/production/long_video_pipeline.py	kling-3.0-pro
conversational	app/production/conversational_pipeline.py	sora-2-pro

Deep Links

• app/video_models.py - Model configs (3 model)
• app/video_styles.py - 10 görsel stil
• app/sora_helper.py - Sora + smart selection
• app/kling_helper.py - Kling via fal.ai
• context/reels-prompts.md - Prompt examples