Open-source tool batch-generates and self-checks images on a laptop, with Gemma agreeing 64% of the time

A developer built Shotfill, an MIT-licensed Python tool that takes a line-by-line shot list and generates each image locally on an Intel laptop without a dedicated GPU. The system draws every shot with open models, uses Gemma 4 to check each result, and redraws the ones that miss. Shots it cannot fix are collected into a short list for the user to review later. The author tested it on a 16-image narrated-video list across three runs, with passes ranging from 5 to 13 of 16 as the code changed. The runs took between 50 and 96 minutes, and the author notes the experiments were not controlled benchmarks.