Transformer-based advertisement image layout generation and object fidelity optimization
Dandan Xu, Kun Hu, Linlu Cai · 2026
Read Transformer-based advertisement image layout generation and object fidelity optimization by Dandan Xu, Kun Hu, Linlu Cai (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0352418.txt.
Description
Dandan Xu, Kun Hu, and Linlu Cai generate advertisement images that obey a planned layout. Their 2026 study builds a two-stage framework. Layout decides where the eye lands, and with it whether the product registers. Text-to-image systems handle simple scenes well but garble complex ones. Stage one turns a written brief into a full layout plan. A transformer generator plans background and objects together, learning long-range links between words and regions. Earlier systems planned objects alone.
Stage two paints the image inside that plan. A fine-grained interaction step keeps text, layout, and pixels in constant conversation. Object fidelity, whether the product itself survives generation intact, is the named optimisation target. Advertisers cannot use a beautiful image whose bottle is warped. The framework treats layout as a contract: plan the whole scene first, then render it faithfully enough to sell.
Catalog record
| CREATOR | Dandan Xu, Kun Hu, Linlu Cai |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63810 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |