Ultra Realistic Media

ULTRA REALISTIC MEDIA GENERATION - TRAINING SKILL

When to Use

Trigger phrases:

• "ultra realistic media"
• "Help me with ultra realistic media"

Use cases:

• When the task matches this skill's domain expertise

When NOT to use:

• For tasks outside this skill's scope

Overview

Train AI agents to generate ultra realistic images and videos using multiple state-of-the-art providers: NVIDIA Flux, BytePlus Seedance, Grok Imagine, and Gemini AI.

Goal: Create photorealistic, cinematic content that's indistinguishable from real photography/video.

🎯 CORE CONCEPTS

Ultra Realism Standards

🛠️ TOOLBOX: 4 PROVIDERS

1. NVIDIA Flux (Image) 🎨

• Best for: Ultra realistic portraits, product photography
• Resolution: 1024×1024, 1024×1792 (9:16)
• Quality: Photorealistic faces, textures
• Cost: ~$0.004 per image
• Strength: 🟢🟢🟢🟢⚪ Realism

2. BytePlus Seedance (Video) 🎬

• Best for: Product showcases, TikTok content
• Duration: 5-60s clips
• Resolution: 704×1248 (9:16), 1024×1024
• Quality: Smooth motion, cinematic
• Cost: ~$0.026 per 5s clip (lite)
• Strength: 🟢🟢🟢⚪⚪ Realism

3. Grok Imagine (Video) 🎥

• Best for: Cinematic shots, dramatic lighting
• Duration: 6-10s clips
• Resolution: High-deff (proprietary)
• Quality: Cinematic color grading, audio sync
• Cost: Super Grok subscription
• Strength: 🟢🟢🟢🟢⚪ Cinematic quality

4. Gemini AI (Product Posing) 👗

• Best for: E-commerce product models, fashion
• Resolution: 9:16 optimal
• Quality: Professional product photography
• Cost: Free (with account)
• Strength: 🟢🟢⚪⚪⚪ Product integration

📚 PROVEN PROMPT FORMULAS

Ultra Realistic Portrait Formula

BASE: [Subject description], [age] years old, [ethnicity]
LIGHTING: [lighting setup], [time of day], [weather]
DETAILS: skin texture, pores, freckles, wrinkles at [age]
CAMERA: 85mm lens, f/1.8, bokeh, natural depth of field
EXPRESSION: [emotion] expression, looking at [direction]
SETTING: [environment], [time period], [cultural context]

Example:

Indonesian woman, 28 years old, Javanese ethnicity
soft natural window light, golden hour, clear weather
visible skin pores, subtle laugh lines, natural complexion
85mm portrait lens, f/1.8, shallow depth of field, bokeh background
gentle smile, looking directly at camera
modern Jakarta cafe, afternoon, 2026, casual contemporary

Ultra Realistic Product Shot Formula

BASE: [product name], [material], [color], condition: new/pristine
LIGHTING: studio three-point lighting, soft diffusion
TEXTURES: [material textures visible], surface reflections
DETAILS: product name visible, professional composition
SETTING: [solid/gradient background], professional product photography style
POST-PRODUCTION: slight color grading, professional retouching

Example:

iPhone 15 Pro, titanium frame, natural titanium color, mint condition
Three-point studio lighting, soft diffused fill, rim light from left
Brushed titanium texture visible, glass reflection on camera module
"iPhone 15 Pro" text visible on back, clean 45-degree side profile
Slate gray gradient background, Apple product photography aesthetic
Slight warmth color grading, subtle highlight boost, commercial quality

Cinematic Story Video Formula

SCENE: [setting], [time of day], [weather/atmosphere]
ACTION: [camera movement], [subject action], [shot size]
MOOD: [emotional tone], [pacing], [color grade]
AUDIO: [ambient sound], [music style], [dialogue if any]
CINEMATIC: film grain, lens flare, dolly movement, slow-motion option

Example:

Jakarta skyline at sunset, golden hour light, thin wispy clouds
Slow dolly camera movement forward revealing cityscape, establishing shot
Nostalgic hopeful mood, slow pacing, warm orange-cyan color grade
City ambient hum, gentle orchestral swell, no dialogue
Film grain overlay, lens flare from setting sun, cinematic quality

🏆 STEP-BY-STEP TRAINING PROGRAM

Phase 1: Image Excellence (Days 1-3)

Day 1: Master NVIDIA Flux Basics

# Generate 20 portrait variations
fluent prompts:
1. Ultra realistic portrait - Indonesian female, 25, natural light
2. Ultra realistic portrait - Indonesian male, 30, studio lighting
3. Ultra realistic portrait - elderly Javanese woman, 60, warm light

# Evaluate: skin texture realism, lighting quality, composition

# Goal: 90% indistinguishable from real photos

Day 2: Product Photography

# Generate 20 product shots
products: phones, watches, jewelry, cosmetics, food

# Prompt focus: material textures, reflections, lighting

# Goal: Commercial-ready product images

Day 3: Advanced Techniques

# Multi-layered compositions
# Environmental portraits (person + setting)
# Dynamic lighting (backlight, rim light, three-point)
# Post-processing simulation in prompts

# Goal: Cinematic stills that tell stories

Phase 2: Video Mastery (Days 4-6)

Day 4: BytePlus Seedance Basics

# Generate 10 short clips (5-10s)
topics: product showcase, lifestyle, environment

# Focus on: smooth motion, natural pacing, loopable

# Evaluate: motion quality, visual consistency, cinematic feel

Day 5: Grok Imagine Cinematic Shots

# Generate 8 cinematic clips (6-10s)
focus: dramatic lighting, camera movement, color grading

# Goal: Movie-quality short scenes

Day 6: Multi-Provider Hybrid

# Generate stills with Flux, animate with Seedance
# Create storyboards with Flux video, enhance with Grok

# Goal: Best of both worlds - realistic + cinematic

Phase 3: Integration (Days 7-10)

Days 7-8: Content Pipelines

# Pipeline 1: TikTok Product Promos
  - Product shots (Flux)
  - Smooth camera pans (Seedance)
  - UGC-style captions

# Pipeline 2: Instagram Lifestyle
  - Portrait shots (Flux)
  - Slow-motion lifestyle (Grok)
  - Aesthetic feed posts

# Pipeline 3: YouTube Shorts
  - Story scenes (Flux + Seedance)
  - Cinematic intros (Grok)
  - Compelling hooks

Days 9-10: Real-World Projects

# Project 1: Complete product launch video
# - Static product shots (Flux)
# - 360-degree rotation (Seedance)
# - Lifestyle integration (Grok)
# - Full promotional video ( stitched with FFmpeg)

# Project 2: Personal branding content
# - Headshot series (Flux)
# - Reel stories (Seedance)
# - BTS footage style (Grok)
# - 30-60s brand intro

# Goal: Commercial-ready assets

🎯 QUALITY CHECKLIST

Image Realism Checklist

• Skin pores visible (portraits)
• Natural eye reflections
• Accurate skin tones/color reproduction
• Proper shadow directionality
• No AI artifacts (extra fingers, distorted faces)
• Natural lighting behavior
• Material textures visible (products)
• No plastic/shiny skin look
• Natural expression/pose
• Appropriate background depth

Video Realism Checklist

• Natural motion (no jerky movements)
• Smooth camera movements
• Proper frame rate consistency
• Natural object behavior (physics)
• Appropriate pacing for content
• No glitching or frame dropping
• Cinematic depth of field (when applicable)
• Color grading consistency
• Audio sync accuracy
• No AI motion artifacts

📊 ADVANCED PROMPTING STRATEGIES

Negative Prompting (What NOT To Generate)

# NVIDIA Flux negative prompts
negative_common = """
cartoon, anime, illustration, 3D render,
uncanny valley, distorted face, extra fingers,
plastic skin, oversaturated, flat lighting,
artificial, blurry, low quality
"""

# Per-use negative prompts
portrait_negative = negative_common + """
old wrinkles (for young subjects), harsh shadows,
flash photography, studio strobe on face
"""

product_negative = negative_common + """
floating product, white background cutout visible,
reflection artifacts, textureless surface,
stock photo aesthetic, generic stock
"""

Prompt Chaining for Consistency

# Step 1: Generate base portrait
prompt_1 = "Indonesian woman, 25, studio lighting..."

# Step 2: Generate variations with consistent features
prompt_2 = prompt_1 + ", slight smile, looking right"
prompt_3 = f"{prompt_1}, laughing, looking up"
prompt_4 = f"{prompt_1}, serious expression, profile view"

# Extract consistent features:
# facial structure, hair style, eye color, skin tone

Iterative Refinement Technique

# Iteration 1: Base prompt
initial_result = generate(base_prompt)

# Iteration 2: Add detail based on critique
refined_prompt = f"{base_prompt}, [specific improvement area]"

# Iteration 3: Fine-tune final output
final_prompt = f"{refined_prompt}, [final tweak]"

# Goal: Converge on ultra-realistic output in 3-5 iterations

🚀 COMMERCIAL APPLICATIONS

E-Commerce Photography

# Complete product shoot automation
for product in catalog:
  # 1. Product isolated shots (Flux)
  isolate_shot = flux.generate(
    prompt=product_isolated_prompt(product),
    negative="white background, stock vibe"
  )

  # 2. Product in context (Flux)
  context_shot = flux.generate(
    prompt=product_in_lifestyle_context(product),
    environment="modern home"
  )

  # 3. 360-degree preview (Seedance)
  rotation_video = seedance.generate(
    prompt=product_rotation_prompt(product),
    style="smooth 360-degree camera orbit"
  )

  # 4. Usage demo (Grok)
  demo_video = grok.generate(
    prompt=f"Person using {product.name} naturally",
    setting="realistic home environment"
  )

Social Media Content

# TikTok UGC-style post
# 1. Portrait with authentic expression (Flux)
portrait = flux.generate(
  prompt=ugc_portrait_prompt(
    subject="Gen Z Indonesian",
    emotion="excited"
  )
)

# 2. Lifestyle moment (Seedance)
lifestyle = seedance.generate(
  prompt="Gen Z scrolling TikTok on phone, authentic",
  shot="POV of someone else looking at them"
)

# 3. Product reveal (Seedance)
reveal = seedance.generate(
  prompt=smooth_reveal_with_product(product),
  camera="push-in reveal"
)

# 4. Compile to 60s TikTok (FFmpeg)
tiktok = ffmpeg.stitch_clip([
  portrait, lifestyle, reveal
], duration=60)

Brand Campaigns

# Full campaign asset generation
campaign_assets = []

# 1. Hero image (Flux)
hero = flux.generate(
  prompt=brand_hero_prompt(company),
  resolution="1024×1792  # 9:16"
)

# 2. Product line showcase (Flux × 10)
products = [flux.generate(
  prompt=product_promo_prompt(p),
  consistency=brand_style_guide
) for p in product_line]

# 3. Lifestyle scenarios (Seedance × 5)
scenarios = [seedance.generate(
  prompt=brand_scenario_prompt(s),
  cinematic=brand_cinematic_style
) for s in lifestyle_scenarios]

# 4. Cinematic brand video (Grok)
brand_film = grok.generate(
  prompt=brand_film_prompt(company),
  length="10s maximum cinematic"
)

⚡ PERFORMANCE TIPS

Speed Optimization

# Parallel generation (Flux allows concurrent)
from concurrent.futures import ThreadPoolExecutor

with ThreadPoolExecutor(max_workers=4) as executor:
  portraits = list(executor.map(
    flux.generate, portrait_prompts_list
  ))

# Batch processing (Flux supports batch)
batch_result = flux.generate(
  prompts=portrait_prompts_list,  # Up to 4 at once
  batch_size=4
)

Quality Cost Tradeoffs

Cost Optimization Strategies

# Strategy 1: Use Flux sparingly (critical shots only)
critical_flux = generate("hero_product_shot", flux)

# Strategy 2: Reuse Seedance clips (loop 5s clips)
looped_seedance = ffmpeg.video_loop(
  seedance_clip, loop_count=12  # 5s → 60s
)

# Strategy 3: Batch prompts (generate 4 at once)
batch_flux = flux.generate_batch([
  "shot 1", "shot 2", "shot 3", "shot 4"
])

# Savings: ~40% cost vs individual generation

🧪 TESTING & VALIDATION

A/B Testing Framework

# Test prompt variations against realism metrics
test_variations = [
  ("Ultra realistic portrait", base_prompt),
  ("Ultra realistic + lighting", f"{base_prompt}, studio lighting"),
  ("Ultra realistic + details", f"{base_prompt}, skin texture visible"),
]

for label, prompt in test_variations:
  image = flux.generate(prompt)
  score = evaluate_realism(image)
  results.append((label, score))

# Compare and iterate on best performers

Realism Evaluation Metrics

def evaluate_realism(image):
  scores = {
    "skin_pore_visibility": check_pores(image),
    "lighting_naturalness": check_lighting(image),
    "color_accuracy": check_colors(image),
    "absence_artifacts": check_no_artifacts(image),
    "overall_realism": human_rating(image)
  }
  return scores

# Benchmark: 90%+ overall = ultra realistic

📈 PROGRESS TRACKING

Weekly Goals

Week 1: Master Flux portraits (90% realistic) Week 2: Master Flux products (commercial quality) Week 3: Master Seedance motion (smooth, cinematic) Week 4: Master Grok cinematic shots (film quality) Week 5: Build first complete pipeline product Week 6: Optimize for speed and cost Week 7: Create 50+ commercial assets Week 8: Realize ultra realistic standard

Portfolio Milestones

• 20 ultra realistic portraits (human judges 90%+ as real)
• 50 commercial product shots (e-commerce ready)
• 30 cinematic video clips (6-10s, film quality)
• 10 complete product videos (60s+)
• 5 brand campaign asset suites (full campaign)
• 1 viral TikTok (100K+ views) using AI assets

🎯 FINAL ASSESSMENT

Ultra Realism Certification Criteria

Images:

• 100+ images generated across categories
• 90%+ rated as indistinguishable from real
• Consistent quality across subject types
• Efficient generation pipeline (<5s per image)

Videos:

• 50+ clips generated (6-60s)
• 90%+ rated as cinematic quality
• Seamless motion, no glitches
• Smooth multi-provider integration

Pipeline:

• Automating content generation at scale
• Cost-optimized production ($0.10 or less per asset)
• Quality control checks in place
• Real-world application delivered

🚀 NEXT STEPS

1. Start Training: Begin Phase 1 (Image Excellence)
2. Build Portfolio: Generate 100+ images, 50+ videos
3. Validate Quality: Get human feedback on realism
4. Optimize Pipeline: Reduce costs, improve speed
5. Apply Commercially: Start generating for real business
6. Iterate: Monthly improvements based on client feedback

Remember: Ultra realistic generation is a continuum, not a destination. The more you practice, the better you get at crafting prompts and understanding AI limitations.

Consistency + Quality = Ultra Realistic Results. 🎯🖼️🎬

For reference: See content-generator/SKILL.md (videos), gemini-image-generator/SKILL.md (products), grok-video-generation/SKILL.md (cinematics)

How to Use

1. Define content goal (traffic, engagement, conversion, brand awareness)
2. Research target audience pain points and search intent
3. Generate content using appropriate AI tools
4. Edit and humanize output for authenticity
5. Optimize for target platform (SEO, hashtags, format)
6. Schedule and distribute across channels
7. Measure performance and iterate

When NOT to Use

• Task is about content strategy, not creation (use strategy skills)
• Task is about content distribution (use distribution skills)
• You need to analyze content performance (use analytics skills)
• Task is about content moderation (use moderation tools)
• You don't have content guidelines
• Task requires domain expertise (consult experts)

Red Flags

• AI-generated content sounds robotic: Always run through humanizer before publishing
• Engagement dropping week-over-week: Content fatigue or algorithm change — vary formats
• Duplicate content across platforms: Adapt content per platform, don't just cross-post
• No content calendar: Sporadic posting kills audience retention
• Ignoring analytics: Content without measurement is just publishing, not marketing

Verification

• Check readability score (target grade 8 or below for general audiences)
• Verify all images have alt text and proper dimensions per platform
• Confirm links work and point to correct destinations
• Test video/audio quality before publishing
• Validate content renders correctly on mobile devices

Process

1. Analyze the task requirements
2. Apply domain expertise
3. Verify output quality

Anti-Rationalization Table