SuperAPI AI Directory

acos-visual-gen

Design & creation

View repository

Updated 10-15 21:47

About this skill

ACOS Visual Generator

What This Skill Does

Creates research-grounded images by:

  1. Researching the topic via web search
  2. Extracting key facts for accuracy
  3. Constructing a detailed image prompt
  4. Generating via Nano Banana MCP (Gemini 2.5 Flash Image)

Prerequisites

  • Nano Banana MCP server configured
  • WebSearch tool available

Quick Start

User: "Create a 3D infographic about quantum computing"

Process:

  1. WebSearch → Extract key facts
  2. Construct prompt with style instructions
  3. mcp__nanobanana__generate_image()
  4. Return image with sources

---

## Visual Styles

| Style | Description |
|-------|-------------|
| `standard` | Clean scientific illustration, modern, professional |
| `minimalist` | Bauhaus, flat vector, 2-3 colors, negative space |
| `photorealistic` | Cinematic lighting, 8K, detailed textures |
| `3d` | Dark Premium 3D — navy background, glass/chrome surfaces, neon accents. NEVER claymorphic |
| `technical` | Da Vinci notebook, ink sketch, annotations |
| `futuristic` | Cyberpunk HUD, neon, holographic |
| `vintage` | 19th century lithograph, sepia, engraving |
| `cartoon` | Educational comic, vibrant, cel-shaded |

## Audience Levels

| Level | Target |
|-------|--------|
| `elementary` | Ages 6-10, bright, simple, fun icons |
| `highschool` | Standard textbook, clean, accurate |
| `college` | Academic journal, high detail, data-rich |
| `expert` | Technical schematic, extremely dense |

---

## Step-by-Step Process

### Step 1: Research
```javascript
WebSearch("{topic} facts key information 2026")

Extract 3-5 verifiable facts.

Step 2: Construct Prompt

Create a {aspect_ratio} {style_description} infographic about {topic}.

VISUAL STYLE: {style_instruction}
AUDIENCE: {audience_instruction}

INCLUDE:
- {fact_1_visualized}
- {fact_2_visualized}
- {fact_3_visualized}

COMPOSITION:
- Clear layout with visual hierarchy
- Text should be large and legible
- Include labels and annotations

Step 3: Generate

mcp__nanobanana__generate_image({
  prompt: "{constructed_prompt}",
  aspect_ratio: "16:9",
  model_tier: "pro",
  enable_grounding: true,
  thinking_level: "high",
  resolution: "high"
})

Step 4: Present Results

Return:

  • Generated image
  • Research sources
  • Key facts included
  • Modification options

Style Instructions (Copy These)

Standard:

High-quality digital scientific illustration. Clean, modern, highly detailed. Professional color palette.

3D Dark Premium:

Dark Premium 3D visualization. Navy background (#0f172a), glass/chrome surfaces with metallic reflections, neon accent glows (cyan #43BFE3, purple #AB47C7, gold #F59E0B). 3+ depth layers, studio lighting, professional architectural render quality. NEVER claymorphic, NEVER pastel, NEVER clay texture.

Technical:

Da Vinci Notebook style. Ink on parchment sketch, handwritten annotation style, rough but accurate lines, technical precision.

Futuristic:

Cyberpunk HUD. Glowing neon blue/cyan lines on dark background, holographic data visualization, 3D wireframes.

Example Usage

User: "Visualize how neural networks learn"

Research: WebSearch finds key facts about backpropagation, gradient descent, layers, activation functions.

Prompt:

Create a 16:9 Dark Premium 3D infographic about how neural networks learn.

VISUAL STYLE: Navy background (#0f172a), glass/chrome surfaces, neon accents...

INCLUDE:
- Input layer receiving data
- Hidden layers processing with weights
- Backpropagation arrows showing error correction
- Output layer with predictions

Generate: Call Nano Banana MCP with grounding enabled.


Integration

This skill integrates with:

  • /infogenius command in ACOS
  • Nano Banana MCP for image generation
  • WebSearch for factual grounding
  • visual-creation skill for quality-gated workflow

Part of the Agentic Creator OS visual generation pipeline.