Auto Ad Generator

Generate professional advertisement posters with AI-powered backgrounds and product compositing.

依赖声明 (Dependencies)

本 Skill 需要以下外部服务和工具：

必需的二进制文件

• python3 - Python 3.8+
• dreamina - Dreamina CLI (通过 curl -s https://jimeng.jianying.com/cli | bash 安装)

必需的环境变量

• DREAMINA_API_KEY - Dreamina API 密钥
• REMOVE_BG_API_KEY - remove.bg API 密钥（可选，用于背景移除）

可选 API 服务

• DALL-E / Midjourney / Stability AI - 用于 AI 图像生成
• Dreamina/即梦 - 主要 AI 生成后端
• remove.bg - 产品图背景移除

Overview

This skill supports multiple industries and platforms:

• Industries: Automotive, cultural tourism, fragrance/beauty, tea, recruitment, public welfare
• Platforms: WeChat official accounts (21:9), Xiaohongshu (3:4), airport displays (16:9/9:16), lightboxes
• Backends: PIL (local/free) or Dreamina/即梦 CLI (AI-powered, high quality)

Quick Start

Generate an Ad

# Using Dreamina AI backend
python main.py --backend dreamina \
  --car ./product.jpg \
  --brand "Brand Name" \
  --subtitle "Product Tagline" \
  --slogan "Marketing Slogan" \
  --platform xiaohongshu \
  --style premium \
  --output ./output

# Using PIL (local, free)
python main.py --backend pil \
  --car ./product.jpg \
  --brand "Brand Name" \
  --output ad.jpg

Interactive Mode

python main.py
# Follow prompts to select backend, platform, and style

Core Workflow

1. Analyze Input

When user provides car images and requests:

• Identify car type: sedan, SUV, MPV, etc.
• Extract key features: color, design highlights, target audience
• Determine ad style: luxury, sporty, family-friendly, tech-focused

2. Gather Requirements

Ask the user (unless provided):

1. Brand name and model?
2. Main headline/slogan?
3. Subtitle/description?
4. Target audience? (young professionals, families, etc.)
5. Celebrity endorser image? (optional)
6. Preferred color scheme? (or auto-detect from brand)

3. Generate Ad Components

Background Generation

Use image generation to create gradient background:

Prompt template:
"Premium gradient background for car advertisement, 
{primary_color} to {secondary_color} smooth gradient, 
subtle light rays, luxury automotive aesthetic, 
minimalist, high-end commercial photography style, 
no text, no car, clean background only"

Car Subject Enhancement

• Clean up the car image (remove background if needed)
• Enhance lighting and reflections
• Position car at 3/4 front angle or side profile

Typography Layout

Layout Structure (1080x1920 vertical):
┌─────────────────┐
│   HEADLINE      │ ← 40-60px, bold, white or contrast color
│   subtitle      │ ← 20-28px, lighter weight
│                 │
│   [CAR IMAGE]   │ ← Main visual, 60% of frame
│   [Celebrity]   │ ← Optional, overlapping or beside
│                 │
│   Slogan        │ ← Bottom area
│   Logo          │ ← Corner placement
└─────────────────┘

4. Style Reference: Li Auto Aesthetic

Color Palettes:

• Premium: Deep blue → Purple gradient (#1a237e → #7c4dff)
• Warm: Orange → Pink gradient (#ff6b35 → #f7931e)
• Cool: Teal → Cyan gradient (#00897b → #00bcd4)
• Dark: Black → Deep gray with subtle blue tint

Typography:

• Headline: Bold, condensed sans-serif
• Subtitle: Light weight, generous letter-spacing
• Slogan: Italic or script for emotional touch

Lighting:

• Soft, diffused key light
• Subtle rim light on car edges
• Gradient background with light source from top

5. Execution Steps

# Pseudo-code for skill execution
def generate_car_ad(car_image, params):
    # Step 1: Analyze car
    car_analysis = analyze_image(car_image)
    
    # Step 2: Generate background
    background = generate_image(
        prompt=build_background_prompt(params['style']),
        size="1024x1536"
    )
    
    # Step 3: Process car image
    car_processed = remove_background(car_image)
    car_enhanced = enhance_lighting(car_processed)
    
    # Step 4: Composite
    composite = overlay_car_on_background(
        background, 
        car_enhanced,
        position="center-bottom",
        scale=0.7
    )
    
    # Step 5: Add text
    final = add_typography(
        composite,
        headline=params['headline'],
        subtitle=params['subtitle'],
        slogan=params['slogan'],
        font_style=params['style']
    )
    
    # Step 6: Add logo
    if params.get('logo'):
        final = overlay_logo(final, params['logo'])
    
    return final

6. Quality Checklist

Before presenting to user:

• Car is the clear focal point
• Text is readable against background
• Color harmony between car and background
• Professional lighting on car
• Brand/logo placement is subtle but visible
• Overall composition follows rule of thirds

Example Outputs

Example 1: Li Auto Style SUV Ad

Input: SUV image, "理想i6", "新形态纯电五座SUV"
Output: Purple-blue gradient, car at 3/4 angle, large white text, celebrity placement

Example 2: Sporty Sedan Ad

Input: Sports sedan, "P7", "纯粹驾驶乐趣"
Output: Dark background with orange accent lighting, dynamic angle, bold typography

Example 3: Family MPV Ad

Input: Minivan, "MEGA", "全家人的幸福空间"
Output: Warm gradient, spacious composition, friendly tone, emphasis on interior space

Tools & Scripts

Required Tools

• Image generation (DALL-E, Midjourney, or local SD)
• Image editing (remove.bg API or local model)
• Text overlay (PIL/Pillow or similar)
• Image composition (layer blending)

Bundled Scripts

• scripts/generate_background.py - Generate gradient backgrounds
• scripts/composite_ad.py - Layer car, background, text
• scripts/typography.py - Add professional text layout

Guidelines

Do

• Match background color to car's personality (sporty=warm, luxury=cool)
• Keep text minimal - one headline, one subtitle max
• Ensure car lighting matches background light source
• Use high-resolution source images (min 1024px width)

Don't

• Overcrowd with too much text
• Use clashing colors (unless intentional for contrast)
• Place car too small in frame (should be 50-70% of composition)
• Ignore the brand's existing visual identity

Edge Cases

• No car image provided: Ask user to upload, or generate concept car
• Low resolution input: Upscale first, or use as thumbnail/concept only
• Multiple cars: Focus on hero car, use others as supporting elements
• Specific brand requirements: Follow brand guidelines over Li Auto style

Reference Materials

See references/ directory for:

• li_auto_examples.md - Analysis of Li Auto ad patterns
• color_palettes.json - Pre-defined gradient combinations
• typography_guide.md - Font pairing recommendations
• composition_templates/ - Layout reference images

Dreamina (即梦) Integration

This skill supports Dreamina CLI for AI-powered background generation with higher quality results.

Prerequisites

# Install Dreamina CLI
curl -s https://jimeng.jianying.com/cli | bash

# Login (required before use)
dreamina login --headless

Two Backend Modes

Mode 1: PIL (Local, Free)

Uses Python PIL to generate simple gradient backgrounds.

• Pros: No API cost, instant, offline
• Cons: Basic quality, limited styles

python main.py --backend pil --car image.jpg --brand 理想 --model i6

Mode 2: Dreamina (AI-powered)

Uses Dreamina's text2image for professional AI backgrounds.

• Pros: High quality, diverse styles, professional aesthetics
• Cons: Consumes credits (~10-50 per image)

python main.py --backend dreamina --car image.jpg --brand 理想 --model i6 \
  --platform xiaohongshu --style premium

Platform Presets

Style Presets

Complete Workflow with Dreamina

# Step 1: Generate AI background
python main.py --backend dreamina \
  --car ./assets/suv.png \
  --brand 理想 \
  --model i6 \
  --subtitle "新形态纯电五座SUV" \
  --slogan "理想，就是活成自己喜欢的样子" \
  --platform xiaohongshu \
  --style premium \
  --output ./output

# Step 2: Check generation status
dreamina query_result --submit_id=\x3Cid_from_step1>

# Step 3: Download result and composite (manual or scripted)
# Step 4: Add typography using composite_ad.py

Credit Management

# Check remaining credits
dreamina user_credit

# Typical consumption:
# - text2image (2k): ~10-20 credits
# - text2image (4k): ~30-50 credits
# - image_upscale: ~20-40 credits

Advanced: Image-to-Image

Use existing image as base for style transfer:

from scripts.dreamina_backend import image_to_image

result = image_to_image(
    prompt="luxury car advertisement style, premium gradient background",
    image_path="existing_car_shot.jpg",
    ratio="16:9"
)