Data Preparation Guide

1. Data Preparation Guide

Preparing Brand-Approved Text & Visual Metadata for AI Creation

High-quality AI outputs start with high-quality inputs. This guide helps students assemble, clean, and structure text and visual data so it can be used effectively and ethically across generative AI platforms. Proper data preparation ensures consistency, creativity, and professional-grade results—especially when working with multimodal tools that combine text and imagery.

A. Assemble Brand-Approved Text Assets

Before generating anything, students should compile a clean, approved text dataset that reflects the brand, project, or narrative they are working on.

Recommended Text Assets

• 1. Brand taglines and value propositions

  2. Product or project descriptions

  3. Mission statements or positioning language

  4. Customer reviews or testimonials (real or simulated, clearly labeled)

  5. Tone-of-voice notes (luxury, playful, academic, cinematic, etc.)

Best Practices

• 1. Remove duplicates, outdated messaging, and conflicting language

  2. Keep tone consistent across all text inputs

  3. Use plain text (no emojis, extra formatting, or markup unless intentional)

  4. Clearly label fictional or AI-generated reviews

Example (Cleaned Text Block)

Brand Tone: Modern, calm, intelligent, nature-forward  

Key Descriptors: intentional, refined, sustainable, human-centered  

Tagline: Designed with intelligence. Guided by purpose.  

Product Description: A lightweight, adaptive solution built for everyday use, balancing innovation with simplicity.  

Customer Insight: “I appreciate how this feels thoughtful rather than over-designed.”

Suggested Platforms

• 1. ChatGPT (text refinement, tone alignment, dataset cleanup)

  2. Google Docs / Notion (collaborative text curation)

  3. Adobe Firefly (when pairing text with design concepts)

B. Clean & Structure Text for Prompt Use

Once text is assembled, it must be structured for prompt engineering, not pasted randomly into AI tools.

Best Practices

• 1. Group text by function: brand voice, features, emotional cues

  2. Keep prompt inputs concise but intentional

  3. Avoid mixing multiple tones in a single prompt

  4. Use bullet-style inputs for clarity

Example (Prompt-Ready Text Structure)

Use the following brand voice:

• Calm, intelligent, modern

• Avoid hype or exaggerated claims

Incorporate these themes:

• Innovation through simplicity

• Human + AI collaboration

• Long-term value

Suggested Platforms

• 1. ChatGPT (prompt testing and iteration)

  2. Claude (longer narrative or structured logic prompts)

C. Prepare Visual Metadata for Multimodal Models

Visual AI tools perform best when paired with clear, descriptive metadata that explains style, mood, composition, and intent.

Recommended Visual Metadata Tags

• 1. Subject (e.g., woman, cityscape, interface, environment)

  2. Style (cinematic, editorial, minimal, surreal, painterly)

  3. Lighting (soft sunset, high contrast, studio light)

  4. Mood (calm, futuristic, emotional, bold)

  5. Camera / Composition (wide shot, close-up, aerial, symmetry)

Best Practices

• 1. Describe visuals as if explaining them to a designer

  2. Avoid vague terms like “cool” or “nice”

  3. Use consistent descriptors across related images

  4. Keep metadata reusable across prompts

Example (Visual Metadata Block)

Visual Style: Cinematic, editorial realism  

Lighting: Soft sunset, diffused light, pastel tones  

Mood: Calm, poetic, aspirational  

Composition: Wide-angle, centered subject, depth of field  

Motion Cue: Implied slow movement, flowing elements

Suggested Platforms

• 1. Midjourney (visual generation using metadata-driven prompts)

  2. ImagineArt (style exploration and refinement)

  3. Adobe Firefly (design-safe visual experimentation)

D. Multimodal Prompt Assembly (Text + Visual)

When combining text and visuals, students should think in layers, not single sentences.

Best Practices

• 1. Start with intent (what the image or output should communicate)

  2. Add structured brand text

  3. Add visual metadata

  4. Add motion or interaction cues if needed

Example (Multimodal Prompt)

Create a cinematic visual that reflects a calm, intelligent brand.

Use a modern, minimal tone with soft lighting and poetic realism.

A human figure interacts naturally with the environment.

Mood is aspirational, not promotional.

Suggested Platforms

• 1. Sora (cinematic video storytelling)

  2. Midjourney + Runway (image-to-video workflows)

  3. ChatGPT (prompt optimization and narrative clarity)

E. Ethical & Academic Guidelines

All data used in the Touro AI Gallery must meet ethical and academic standards.

Requirements

• 1. Do not upload private, sensitive, or proprietary data

  2. Clearly label fictional or simulated content

  3. Credit inspiration sources where relevant

  4. Include prompts alongside final outputs

This transparency reinforces that AI outputs reflect student thinking, not automation alone.