Skip to content

Image Generation Analysis

Analysis pipeline for studying image generation patterns, regeneration behavior, and prompt effectiveness (April–June 2026).

Scripts

Script Purpose
dump_image_gen_jobs.py Pulls all image generation jobs from MongoDB (renderboard.assetGenJobs) for a date range and saves to JSON
image_gen_insights.py Computes per-user generation counts and regeneration metrics (mean/median/max cluster sizes) using MinHash/LSH
prompt_feature_extraction.py Extracts 15 structural features from each unique prompt via GPT, correlates features with download success rate
compare_sketch_vs_manual.py Compares platform-generated (generateFromSketch) prompts vs manual user prompts — structural and qualitative

Setup

pip install pymongo datasketch openai

Usage

# 1. Dump data from MongoDB
python dump_image_gen_jobs.py

# 2. Regeneration analysis
python image_gen_insights.py

# 3. Feature extraction (uses GPT API, cached after first run)
python prompt_feature_extraction.py

# 4. Platform vs manual comparison
python compare_sketch_vs_manual.py