v1.5.0

[1.5.0] - 2026-07-23

Added

• search_images now returns a sizes map on every photo (the delivered width/height in px for each src variant, accounting for retina dpr) plus a size_guide field explaining how to match a variant's width to the target slot's rendered width, so the model picks src.large/src.large2x for a hero, src.medium for a card/thumb, and src.tiny for an avatar instead of guessing — avoiding pixelated upscaled images and bloated pages from oversized variants in small slots.

Changed

• The generation guide's IMAGES rule now teaches variant-to-slot size matching (using the new sizes/size_guide data and each photo's native width) instead of the previous bare "large for hero, medium for card" rule of thumb.