Créer des vidéos virales avec l'IA de A à Z

1. Ideation     → Find the viral concept
2. Script       → Write the script with an LLM
3. First Frame  → Generate the starting image (image gen)
4. Video Gen    → Transform the image into video (I2V)
5. Audio        → Voiceover / music (TTS / generation)
6. Metadata     → Title, description, tags, hashtags
7. Upload       → Automated multi-platform publication

## Video Ideation Prompt

You are an expert in viral content on TikTok, YouTube Shorts, and Instagram Reels.

Niche: [your niche]
Audience: [your audience]
Current trends: [observed trends]

Propose 5 short video concepts (15-60 sec) with:
- Hook (first sentence/image)
- One-line concept
- Targeted emotion
- Viral potential (score /10)
- Recommended format (talking head, cinematic, tutorial, storytelling)

openclaw cron add \
  --name "Trend watcher" \
  --cron "0 10 * * 1,4" \
  --tz "Europe/Paris" \
  --session isolated \
  --message "Analyze TikTok and YouTube Shorts trends in the tech/AI niche. Identify 3 popular formats this week. Propose adaptations for our channel. Save to trends.json." \
  --model "sonnet"

## Short Script Structure

### Hook (0-3 sec)
- Shocking phrase or provocative question
- Striking opening image

### Development (3-45 sec)
- Main point
- Visual demonstration/proof
- Twist or surprise

### Conclusion (45-60 sec)
- Call to action
- Tease for the next part
- Last memorable image

## Video Script Prompt

Write a short video script (30-45 seconds) on the following topic:
[SUBJECT]

STRICT output format:

HOOK: [Exact text to display/say in the first 3 seconds]

SCENE 1:
- Duration: [X sec]
- Visual: [Precise description of what's seen]
- Narration: [Voiceover text]
- Screen text: [Text displayed on screen, if relevant]

SCENE 2:
[...]

CTA: [Final call to action]

FIRST_FRAME_PROMPT: [English prompt to generate the starting image]

Rules:
- The hook must create immediate tension or curiosity
- Each scene must have a concrete visual description
- Narration must be natural and rhythmic
- FIRST_FRAME_PROMPT must be compatible with AI image generators

## Good First Frame Prompt

"A young tech entrepreneur sitting at a futuristic holographic desk, 
blue neon lighting, cyberpunk office environment, looking at camera with 
confident expression, dramatic rim lighting, shallow depth of field, 
cinematic composition, 9:16 vertical aspect ratio, photorealistic, 
8k quality"

## Bad First Frame Prompt

"Person at desk with computer"

import requests

def generate_video_kie(image_url, prompt, duration=5):
    """Generate a video via KIE.ai API"""
    response = requests.post(
        "https://api.kie.ai/v1/video/generate",
        headers={"Authorization": f"Bearer {KIE_API_KEY}"},
        json={
            "model": "kling-v1.6",
            "image_url": image_url,
            "prompt": prompt,
            "duration": duration,
            "aspect_ratio": "9:16",
            "mode": "professional"
        }
    )
    task_id = response.json()["task_id"]
    return task_id

## Good I2V Prompt

"Slow camera push in, the character turns head slightly to the right 
and smiles, subtle hair movement from wind, ambient particles floating 
in the air, smooth cinematic motion"

## Bad I2V Prompt

"A person at a desk" (describes the scene, not the movement)

## Character Reference Technique

1. Create a "character sheet" with 3-4 reference images
2. Use these images as references in each generation
3. Maintain a consistent character description prompt

Example of persistent description:
"Sarah, 28 years old, short brown hair with subtle highlights, 
green eyes, light skin, wearing a dark blue tech company hoodie, 
confident posture"

def generate_consistent_character(base_prompt, character_desc, seed=42):
    full_prompt = f"{character_desc}, {base_prompt}"
    return generate_image(
        prompt=full_prompt,
        seed=seed,
        style="photorealistic",
        aspect_ratio="9:16"
    )

def generate_video_metadata(script, platform):
    prompt = f"""
    Generate optimized metadata for {platform}:

    Video script: {script}

    Return in JSON:
    - title: catchy title (< 100 chars)
    - description: SEO-optimized description (150-500 chars)
    - tags: 10-15 relevant tags
    - hashtags: 5-8 trending hashtags
    - thumbnail_text: short text for thumbnail (3-5 words)
    - best_posting_time: optimal posting time
    """
    # ... (AI generation code)