🛰️ Daily AI Frontier
‹ back to 2026-09-04

Editable Visual Design

arXiv cs.CV LLM Agents Junyan Ye, Wei Liu, Dongzhi Jiang, Zichen Wen, HaoDong Li, Zhutao Lv, Jiaxin Lin, Jinhua Yu, Jun He, Zilong Huang, Rui Chen, Weijia Li 2026-09-03
Representative image for Editable Visual Design

TL;DR - Editable Visual Design is a coding-agent framework that combines VLM-guided planning and aesthetic judgment with image generation and native HTML/CSS authoring. It aims to preserve generative models’ visual quality while producing layered, text-correct artifacts that remain directly editable.

  • Uses a VLM as the “creative brain” for interpreting requirements, planning, and evaluating aesthetics.
  • Generates complex visual assets on demand, then assembles them as decoupled layers with native HTML/CSS and real text.
  • Iteratively improves designs through an “imagine first, then act” loop informed by rendered visual feedback.
  • Supports GUI-based dragging and layout adjustments, with validation reported across posters, infographics, and related designs.

view merged work →