Diffusion models like GPT-Image-2 generate expressive visuals but output flattened bitmaps with inaccurate text that cannot be edited layer by layer. Code-based approaches using offer precise layout and decoupled layers but lack aesthetic intuition and struggle to generate complex visual assets. This paper presents a new paradigm that combines a VLM as a "creative brain" for planning and aesthetic judgment with an image generation model as an on-demand "visual world simulator" to produce standalone assets. The agent follows a closed-loop workflow: it imagines a design, generates isolated assets, writes native HTML/CSS, and iteratively refines based on visual rendering feedback. The resulting artifacts have decoupled layers and real text, enabling users to drag and adjust layout on a graphical interface. Validations on posters, infographics, and other scenarios demonstrate that this approach achieves both refined aesthetics and production-grade editability, overcoming the limitations of each prior path.
Coding agents combine generated imagery with editable web-based visual layouts
The proposed workflow uses a vision-language model to plan and critique designs, an image model to create assets, and HTML/CSS to preserve editable layers and text.
Independent
Junyan Ye · Wei Liu · Dongzhi Jiang · Zichen Wen · HaoDong Li · Zhutao Lv · +6 more
Research Digest··2 min read
In one line
A Coding Agent combining a VLM and image generation model produces editable visual designs with refined aesthetics and production-grade editability.
Before you read this
Three ideas this paper builds oncoding agentAn AI-powered software system that autonomously writes, modifies, or extends source code based on high-level instructions.134 papers →vision-language modelA model that handles tasks combining visual and textual information, such as editing images from conversation or reasoning across images and text.216 papers →agent-driven simulationAutonomous agents automatically generate, control, or extend simulations to improve accuracy, interactivity, or performance.32 papers →
Read it as
Ye and colleagues present Editable Visual Design, a system for producing visual designs that remain structurally editable rather than being flattened into a single image.
The authors built a closed-loop design agent around a (VLM), which interprets requirements, plans the composition, writes layout code, and judges rendered results.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.