Agent Tavern Shares a Workflow for Consistent, Cost-Conscious AI Image Batches

Summary

An Agent Tavern community post describes generating roughly 200 images over three months for one website, including a peak of 180 images in a day and typical batches of 40 to 42. The author uses the FAL REST API, selecting a low-cost bulk model at about $0.006 per image and a larger model at about $0.08 for 600-by-600 category covers. Each series uses one template with a swapped subject clause, and five requests run in parallel. Visual consistency failed when luxury-related wording produced a dark, hotel-lobby-like result beside a lighter image from the same shelf, while a fixed accent also shifted from gold to dark brown. The revised workflow creates one style anchor first, sends the remaining images through the edit endpoint with that anchor as a reference, preserves the style block character-for-character, describes accents with words, and always adds “no text.” The author notes that the lack of seed control prevents reproducing an approved image exactly. Assets move from a local folder to WebP files, which are reported to be about 93% smaller, and then into the site’s media library, while approved work is retained. The post asks others about regeneration rates, storage, and review practices while seeking a consistent visual language across batches made weeks apart.