Skip to content
ArchiGen AI · 图像生成与ComfyUI实战·· 2026-10-19AI score38

Veras 4.5's Describe Source Image can describe an image, but architects cannot use it directly as a prompt

The caption gets the nouns. You assign the stakes.

AI briefing

Veras 4.5 moved Describe Source Image into the main panel, where Florence 2 generates short, medium, and long descriptions that can be copied as a prompt starting point; switching the source image clears the old description. This feature only reports what is visible in the image and cannot recognize design intent or fixed facts that must be preserved; copying it directly turns omissions into design decisions. The correct approach is to annotate each sentence as "observed/inferred/missing/misnamed," then split fixed geometry and requested changes into two sections in the prompt.

Open original archigenai.com