SearchGen Studio
Image generators fabricate what they don't know. This one looks it up first — and knows when not to.
Demo Setup
This live demo runs entirely on our own trained stack — with no live web calls at inference time. It illustrates the knowledge boundary: the line between what a model can learn once and what it must look up per request. Try a knowledge-hungry prompt (a 2025 event, a niche mascot, a specific person) and compare the two modes below.
The agentic Gate → Filter → Integrate pipeline: it fires web/image search only when a real knowledge gap exists, keeps the reference that fills that gap, and routes it through language so nothing extraneous leaks in.
The prompt-rewriting baseline: it enriches wording but still generates from stale weights — great for concepts the model already knows, blind to everything past its training cutoff.
Enjoying the demo? Help us out.
@article{wang2026searchgen,
title = {Search Beyond What Can Be Taught: Evolving the Knowledge
Boundary in Agentic Visual Generation},
author = {Wang, Haozhe and Feng, Weijia and Yu, Jinpeng and Liu, Che and
Nie, Ping and Lin, Fangzhen and Liu, Jiaming and Huang, Ruihua and
Lin, Jimmy and Chen, Wenhu and Wei, Cong},
journal = {arXiv preprint arXiv:2607.05382},
year = {2026},
url = {https://arxiv.org/abs/2607.05382}
}