Airtable schema for AI image prompt management

Database schema specification for managing AI image generation workflows. A planning document exploring how to structure prompt libraries across multiple generation platforms.

The Problem

AI image generation workflows need structured tracking of:

• Prompts and their parameters (which vary wildly by platform)
• Generated outputs and their quality
• Iteration history (prompt refinement over time)
• Review workflows for systematic quality assessment

Schema Architecture

Three-table normalized structure with clear relationships:

Prompts → Generated Images → Reviews
   ↓
Prompts (self-referential for iteration tracking)

Platform Coverage

Platform	Key Parameters
DALL-E 3	Quality (standard/hd), Style (vivid/natural)
Midjourney	Stylize, Chaos, Weird, Style refs
Stable Diffusion	CFG scale, Sampling steps, Sampler, LoRA, VAE
Flux	Similar to SD with scheduler options
Leonardo AI	Platform-specific presets
Adobe Firefly	Style/content controls

Prompts Table Highlights

Core fields:

• Prompt text + negative prompt
• Platform & model version
• Style tags (multiple select)
• Universal params: seed, aspect ratio, dimensions

Platform-specific params:

• SD/Flux: CFG scale, steps, sampler, LoRA models, checkpoint
• Midjourney: stylize (—s), chaos (—c), weird, style refs
• DALL-E: quality, style presets

Iteration tracking:

• Self-referential parent/child links
• Iteration number and notes
• What changed between versions

Aggregation Fields

Prompt Effectiveness Score =
  IF(Image Count > 0,
     ROUND(Average Rating * (1 + LOG(Image Count)/5), 1),
     0)

Combines quality (average rating) with quantity (image count) to surface prompts that consistently produce good results.

Storage Considerations

Airtable Team plans: 20GB per base, 5GB max per file. Schema recommends image compression before upload to stay within limits while accommodating high-resolution outputs.