Rug shoppers and interior designers struggle to picture how a specific rug will look in a real room, and static product photos rarely close that gap. Rug & Kilim wanted a way for anyone to see any rug from the catalog placed in their own space, or any space they could describe, in seconds.
02The Solution
+
The studio pairs Algolia instant search (pick any rug from the catalog) with Google's Gemini image models. Shoppers describe a room, upload a photo of their space, or choose a preset, and Gemini composites the selected rug into the scene at the correct real-world scale. A dimension-aware prompt engine reads each rug's size and aspect ratio to place it believably, and the whole thing runs behind a hardened endpoint with model failover and abuse protection.
03Inside the studio
+
Shoppers pick a rug from Algolia-powered catalog search, then describe a room, upload a photo of their space, or choose a preset, all from a single page.The studio: catalog search and scene builder
04Features
+
Algolia powers a live, typo-tolerant search across the full rug catalog, so any product can be staged in seconds.
Type out a room, upload a photo of your own space, or pick a curated editorial preset. Three ways to set the scene.
A prompt engine reads each rug's real size and aspect ratio so it lands at believable scale: runners stay narrow, room-size rugs fill the floor, tapestries hang on the wall.
Layered upload validation and abuse protection keep the unauthenticated generation API safe.
Automatic failover and load management keep generations flowing even when the primary model is overloaded.
Voice-to-text scene descriptions, plus a moderated gallery of approved generations from shoppers around the world.
05Generated results
+
Real outputs from the studio: each rug composited into a described or uploaded room at believable scale, with lighting and perspective matched to the scene.Living room with a stone fireplace
06System Architecture
+
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Withheld
Confidential
Proprietary architecture, kept confidential while the product is live.
Full stack details kept confidential for competitive reasons.
08Key Learnings
+
01The hard part was not calling the image model, it was making it reliable in production. Automatic failover turned an overloaded primary model from a 60-second hang into a seamless switch to a backup.
02Image quality lived in the prompt. Grounding the model in each rug's real-world dimensions was what made placements look believable instead of toy-sized.
03A public, unauthenticated generation endpoint needs real guardrails: layered upload validation, rate limiting, and load management to protect both cost and the server.
04Pairing Algolia instant search with the generator made the flow feel immediate. Users find a rug and see it in their room without ever leaving the page.
nahom endale
