AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards
An arXiv paper introduces AesCode, a method that generates code grounded in a content prompt while using an image generated from that same prompt as aesthetic guidance. It represents the target as a design graph covering the whole canvas so each property is attributable, then derives complementary cross-modal rewards: deterministic verifiers score properties parsable from code and rendering, while a VLM judge assesses non-parsable ones using a sample-specific Visual Graph Rubric tied to graph elements and relations, optimized via reinforcement learning with per-channel normalization. The authors also build a benchmark scoring verifiable correctness and visual quality separately, and report that AesCode-8B and AesCode-32B improve visual quality by 31.1 and 22.4 percentage points over same-scale Qwen3-VL baselines given the same references, reaching scores comparable to closed-source frontier models such as GPT-5.5 and Claude Opus 4.8 (author claims).
It targets a known weakness of code models that cannot see the canvas, pairing image references with attributable rewards to make generated slides, posters and dashboards both prettier and verifiable.