product-photo-batch-pipelinelisted

# Batch Image Pipeline A catalogue is not one image done many times. It's a queue with rate limits, a credit budget, a consistency requirement and a review step — and the review step is the one that gets dropped first and matters most. ## The rate limits | Endpoint | Per minute | |---|---| | `/generations`, `/brand-assets` | **120** | | `/ai-image-generate` | **60** | | `/image-studio` | see plan | | Background tools (`remove_background`, `replace_background`) | **30** | **Background tools are the bottleneck at 30/minute**, and they're usually the first step in a catalogue pipeline. A 300-product cutout job is at least ten minutes of wall clock at the limit. Images are synchronous, unlike video — so the pipeline is a rate-limited loop rather than a job queue with polling. That's simpler, and it means throttling is the only concurrency concern. ## The pipeline ```python import time RPM_BACKGROUND = 30 GAP = 60.0 / RPM_BACKGROUND + 0.3 # margin LOOK = ("On pure white, soft even studio light from above and slightly " "left, neutral white balance, accurate colour, subtle contact " "shadow. Product fills 85% of frame. Product unchanged.") def process(product, log): pid = product["id"] try: cutout = remove_background(image=product["photo"]) time.sleep(GAP) asset = create_brand_asset(type="product", ...) # free shots = {} for angle in ("straight on, front elevation", "three-qua