Renova 如何打造 AI 空間改造功能
How we built Renova's AI room redesigns
Renova 以提示詞驅動的影像編輯模型配合浮水印，建立 AI 空間改造流程，讓用戶點按後數秒內，把房間、房屋外觀、花園或泳池照片改成指定風格的同一空間。提示詞會保留空間結構、建築格局和拍攝角度，多種改造工具共用同一核心步驟。Renova 以同一套程式碼支援 iPhone、Android 和網頁版，並讓同一帳戶及設計記錄在各平台間延續。
Renova takes one photo of a room, a house exterior, a garden or a pool and turns it into the same space redesigned in a chosen style, seconds after you tap. Here's roughly how that pipeline came together.
Keep the photo's geometry, change everything else. Renova is prompts on an image-edit model, plus a watermark. The prompts are written to keep the room's architecture and camera angle, so a "Scandinavian" redesign still has your window in the same wall, at the same size.
One pipeline, many tools. Style redesign, virtual staging (furnishing an empty room), new walls and floors, furniture rearrangement, style transfer from a reference photo, floor-plan-to-3D-rendering, and occasion decorating all share the same core step, just with different inputs and constraints. That made it possible to add a new "tool" without building a new pipeline from scratch each time.
One codebase, three platforms. Renova ships on iPhone, Android and the web from the same codebase, with the same account and the same design history following you between them.
If you work on anything that edits or regenerates real photos rather than generating from scratch, the "keep structure, change material" problem is probably familiar — happy to compare notes in the comments.
來源：DEV Community · dev.to