Based on a brand description or reference image, batch generate 10 professional logo variants, supporting 7 mainstream styles.
Two Entry Modes
Mode A: Text Description Entry
The user describes the brand name, purpose, and style preference → generate the logo.
Mode B: Image Reference Entry
The user uploads a reference image → analyze the image elements → ask about requirements → generate the logo.
Workflow
Phase 1: Requirements Collection
Mode A (Text Description):
- Collect brand information: brand name (Chinese/English), industry/purpose, core concept
- Show 7 style options (read references/logo-styles.md), briefly introduce the characteristics of each style
- Recommend 1-2 most suitable styles based on the brand attributes, and explain the reasons
- Wait for the user to choose a style
Mode B (Image Reference):
- Analyze the image uploaded by the user and extract:
- Graphic elements (shapes/icons/symbols)
- Color scheme (primary color/secondary colors)
- Typography style (font type/arrangement)
- The closest style type
- Emotional tone
- Show the analysis results and ask the user for their intent:
- "Want a logo in a similar style?"
- "Want to keep some elements and improve them?"
- "Want a logo in a completely different direction?"
- Show style options and make recommendations
- Wait for the user to choose
Phase 2: Generate Prompts
After the style is selected, generate prompts for 10 logo variants for the brand:
Variation Strategy(read references/logo-styles.md to get the template):
- Variants 1-2: Primary style, different compositions (icon centered / image and text arranged left and right)
- Variants 3-4: Primary style, different color schemes (light/dark/contrasting colors)
- Variants 5-6: Primary style, different icon/symbol variations
- Variants 7-8: Crossovers of similar styles (e.g., minimalist + line hybrid)
- Variants 9-10: Additional exploration directions (user-specifiable)
Structure of each prompt:
- Style description (from the template prefix of the selected style)
- Specific brand description (name, industry, concept)
- Specific variations in composition/color scheme
- Base instructions:
Professional logo design. Pure white background. High contrast. Scalable vector-style quality. No watermark, no mockup, no background texture. - Chinese instructions (if the logo contains Chinese):
ALL text in the logo MUST use Chinese characters with clean professional Chinese typography. Technical terms and brand abbreviations may remain in English.
Show a summary of all 10 prompts and wait for user confirmation:
## Logo 提示词概要
| # | 变体方向 | 风格 | 配色 | 简述 |
|---|---------|------|------|------|
| 1 | 主风格-构图A | 极简扁平 | 蓝色 | 几何图标居中... |
| 2 | 主风格-构图B | 极简扁平 | 蓝色 | 图文横排... |
| ... |
确认后开始生成（共 10 张 Logo）
Phase 3: Batch Generation
After confirmation, call the script one by one to generate:
python3 scripts/generate_image.py \
--prompt "提示词" \
--output "/Users/ugreen/Documents/obsidian/09image/MMDD-品牌名-logo/01-变体描述.png" \
--api-key "REMOVED_GEMINI_KEY" \
--api-url "https://generativelanguage.googleapis.com" \
--model "gemini-3-pro-image-preview" \
--aspect-ratio "1:1" \
--resolution "2K"
Notes:
- Logos use by default 1:1 square ratio
- Resolution defaults to 2K
- MMDD is the current date
- Use a short English/pinyin brand name (e.g., spaceai, mylogo)
- File naming:
01-风格-变体.png(e.g.,01-flat-icon-center.png） - Interval of 2-3 seconds between each image to avoid rate limiting
- Continue to the next image if one fails, report at the end
Phase 4: Presentation and iteration
- After generation is complete, list all successful Logo paths
- Ask the user:
- "Which directions do you like most?"
- "Do you need to continue iterating based on one direction?"
- "Do you need to adjust the color scheme/shape/text?"
- If the user requests iteration, adjust the prompt based on the selected Logo and regenerate
Save path conventions
/Users/ugreen/Documents/obsidian/09image/MMDD-品牌名-logo/
├── 01-flat-icon-center.png
├── 02-flat-text-horizontal.png
├── 03-flat-dark-theme.png
├── 04-flat-gradient-accent.png
├── 05-flat-alt-symbol.png
├── 06-flat-abstract-shape.png
├── 07-lineart-monoline.png
├── 08-lineart-detail.png
├── 09-explore-badge.png
└── 10-explore-gradient.png
API configuration
| Configuration item | Value |
|---|---|
| API URL | https://generativelanguage.googleapis.com |
| API Key | REMOVED_GEMINI_KEY |
| Model | gemini-3-pro-image-preview |
| Default aspect ratio | 1:1 (square) |
| Default resolution | 2K |
Quick overview of 7 optional styles
| # | Style | One-sentence characteristic | Applicable scenarios |
|---|---|---|---|
| 1 | Minimalist flat | Geometric + solid colors, clean and crisp | Tech, SaaS, internet |
| 2 | Line art | Single line, elegant and refined | Studios, design brands |
| 3 | Letter mark | Creative distortion of initials | Enterprises, personal brands |
| 4 | Badge stamp | Circular border, retro ceremonial feel | Communities, podcasts, education |
| 5 | Mascot icon | Characterized, highly memorable | Games, education, social |
| 6 | Modern gradient | Rich gradients, strong tech feel | AI products, apps, fintech |
| 7 | Handwritten calligraphy | Handwritten style, warm and full of warmth | Blogs, dining, handmade |
For detailed style templates and color reference, see references/logo-styles.md.
Common usage
| User says | Action |
|---|---|
| "Help me design a Logo" | Mode A: Collect description → Recommend style → Generate |
| "Make a Logo, tech style" | Mode A: Directly recommend minimalist/gradient styles |
| "Make a Logo based on this image" | Mode B: Analyze image → Ask intent → Generate |
| "I like the 3rd one, make a few more similar ones" | Iteration: Expand generation based on the selected direction |
| "Change the 5th one to blue" | Iteration: Adjust the color scheme and regenerate |