vapp.py

import os
import time
import json
from typing import Dict, Any, List, Literal
import tempfile
from urllib.request import urlretrieve
from pydantic import BaseModel, constr
from google import genai
from dotenv import load_dotenv

load_dotenv()

GEMINI_API_KEY = os.getenv("GEMENI_KEY", "")

def configure_gemini() -> genai.Client:
    if not GEMINI_API_KEY:
        raise RuntimeError("GEMENI_KEY is not set in environment variables.")
    return genai.Client(api_key=GEMINI_API_KEY)

Timestamp = constr(pattern=r'^\d{2}:\d{2}$')
RangeTimestamp = constr(pattern=r'^\d{2}:\d{2}-\d{2}:\d{2}$')
Score010 = constr(pattern=r'^(?:10|[0-9])\/10$')

class Hook(BaseModel):
    hook_text: str
    principle: str
    advantages: List[str]

class StoryboardItem(BaseModel):
    timeline: Timestamp
    scene: str
    visuals: str
    dialogue: str
    camera: str
    sound_effects: str

class ScriptLine(BaseModel):
    timeline: Timestamp
    dialogue: str

class VideoMetric(BaseModel):
    timestamp: RangeTimestamp
    element: str
    current_approach: str
    effectiveness_score: Score010
    notes: str

class VideoAnalysis(BaseModel):
    effectiveness_factors: str
    psychological_triggers: str
    target_audience: str
    video_metrics: List[VideoMetric]

class TimestampImprovement(BaseModel):
    timestamp: RangeTimestamp
    current_element: str
    improvement_type: str
    recommended_change: str
    expected_impact: str
    priority: Literal["High", "Medium", "Low"]

class AdAnalysis(BaseModel):
    brief: str
    caption_details: str
    hook: Hook
    framework_analysis: str
    storyboard: List[StoryboardItem]
    script: List[ScriptLine]
    video_analysis: VideoAnalysis
    timestamp_improvements: List[TimestampImprovement]

analyser_prompt = """You are an expert video advertisement analyst. Analyze the provided video and give response conforms EXACTLY to the schema below with no extra text or markdown. Populate:
1. **brief** → A concise summary covering visual style, speaker, target audience, and marketing objective.  
2. **caption_details** → Description of captions (color/style/position) or exactly the string `"None"` if not visible.  
3. **hook** →  
   - `"hook_text"`: Exact opening line or, if no speech, the precise description of the opening visual.  
   - `"principle"`: Psychological/marketing principle that makes this hook effective.  
   - `"advantages"`: ARRAY of 3–6 concise benefit statements tied to the ad’s value proposition.  
4. **framework_analysis** → A detailed block identifying copywriting/psychology/storytelling frameworks (e.g., PAS, AIDA). Highlight use of social proof, urgency, fear, authority, scroll-stopping hooks, loop openers, value positioning, and risk reversals.  
5. **storyboard** → ARRAY of 4–10 objects. Each must include:  
   - `"timeline"` in `"MM:SS"` (zero-padded)  
   - `"scene"` (brief)  
   - `"visuals"` (detailed)  
   - `"dialogue"` (exact words; use `""` if none)  
   - `"camera"` (shot/angle)  
   - `"sound_effects"` (or `"None"`)  
6. **script** → ARRAY of dialogue objects, each with `"timeline"` (`"MM:SS"`) and `"dialogue"` (exact spoken line).  
7. **video_analysis** → OBJECT with:  
   - `"effectiveness_factors"`: Key factors that influence effectiveness  
   - `"psychological_triggers"`: Triggers used (e.g., scarcity, authority)  
   - `"target_audience"`: Audience profile inferred  
   - `"video_metrics"`: ARRAY of objects with:  
     - `"timestamp"`: `"MM:SS-MM:SS"`  
     - `"element"`: The aspect being evaluated (e.g., Hook Strategy)  
     - `"current_approach"`: Description of current execution  
     - `"effectiveness_score"`: String score `"X/10"` (integer X)  
     - `"notes"`: Analytical notes  
8. **timestamp_improvements** → ARRAY of recommendation objects with:  
   - `"timestamp"`: `"MM:SS-MM:SS"`  
   - `"current_element"`: Current content of the segment  
   - `"improvement_type"`: Category (e.g., Hook Enhancement)  
   - `"recommended_change"`: Specific recommendation  
   - `"expected_impact"`: Projected effect on metrics or perception  
   - `"priority"`: `"High"`, `"Medium"`, or `"Low"`
⚠️ The output must be strictly matching field names and types, no additional keys, and all timestamps must be zero-padded (`"MM:SS"` for single points, `"MM:SS-MM:SS"` for ranges).
"""

def analyse_video_only(video_path: str) -> Dict[str, Any]:
    client = configure_gemini()
    try:
        video_file = client.files.upload(file=video_path)
        while getattr(video_file.state, "name", "") == "PROCESSING":
            time.sleep(2)
            video_file = client.files.get(name=video_file.name)
        if getattr(video_file.state, "name", "") == "FAILED":
            return {}

        print("Uploaded the video")
        resp = client.models.generate_content(
            model="gemini-2.0-flash",
            contents=[analyser_prompt, video_file],
            config={"response_mime_type": "application/json",
                   "response_schema": AdAnalysis,}
        )
        raw = getattr(resp, "text", "") or ""
        try:
            model_obj = AdAnalysis.model_validate_json(raw)
            return model_obj.model_dump()
        except Exception:
            try:
                return json.loads(raw)
            except Exception:
                return {}
    except Exception:
        return {}

def download_video_to_temp(video_url):
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
    temp_file_path = temp_file.name
    temp_file.close()
    urlretrieve(video_url, temp_file_path)
    print("Downloaded the video")
    return temp_file_path

def get_analysis(video_url):

    file_path = download_video_to_temp(video_url)
    analyses = analyse_video_only(file_path)

    return analyses