Search Beyond What Can Be Taught: Evolving the Knowledge Boundary in Agentic Visual Generation

Source

Source name: arXiv
Source kind: paper
Authors / organization: Haozhe Wang, Weijia Feng, Jinpeng Yu, Che Liu, Ping Nie, Fangzhen Lin, Jiaming Liu, Ruihua Huang, Jimmy Lin, Wenhu Chen, Cong Wei
Venue: arXiv: cs.CV
Published: 2026-07-06
Collected: 2026-07-07
URL: http://arxiv.org/abs/2607.05382v1
PDF: https://arxiv.org/pdf/2607.05382v1
DOI: UNKNOWN

Plain-Language Brief

Researchers have created two datasets called SearchGen-20K and SearchGen-Bench, containing 20,839 prompts across twelve failure categories and twenty-two domains to evaluate the performance of visual generators. They found that these generators struggle with fabricating what they don't know due to their training on fixed corpora. To address this issue, the researchers propose a teach-then-search co-training framework that improves visual generation by discovering a knowledge boundary between what a generator can internalize and what must remain in external context.

What the Document Says

Researchers created SearchGen-20K and SearchGen-Bench datasets with prompts across twelve failure categories and twenty-two domains for evaluating visual generators.
Visual generators perform poorly on these datasets, scoring only 21 to 28 out of 100, as they confidently fabricate what they don't know due to their training on fixed corpora.
The researchers propose a teach-then-search co-training framework for improving visual generation by discovering the knowledge boundary between what a generator can internalize and what must remain in external context.
They found that naive search fails as it retrieves indiscriminately, injecting noise into prompts already handled by the generator.

Key Concepts Mentioned

visual generators
search tools
co-training framework
knowledge boundary

Claims Made by the Document

Visual generators struggle with fabricating what they don't know due to their training on fixed corpora.
The teach-then-search co-training framework improves visual generation by discovering the knowledge boundary between what a generator can internalize and what must remain in external context.

Evidence Used

Visual generators score only 21 to 28 out of 100 on SearchGen-Bench, demonstrating their struggles with fabrication.
The teach-then-search co-training framework produces monotonic improvement in visual generation.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Visual generators excel at rendering, but they confidently fabricate what they do not know. User requests are unbounded, evolving, and deeply long-tailed: new characters, trending entities, post-cutoff events, and more. This world-knowledge bottleneck is structural: generators are trained on fixed corpora, but the visual world is open-ended. We construct SearchGen-20K and SearchGen-Bench, with 20,839 prompts spanning twelve failure categories and twenty-two domains, paired with a pre-executed multimodal SearchGen-Corpus-1M to support offline, reproducible research. On SearchGen-Bench, frontier open generators score only 21 to 28 out of 100, a 40-point collapse invisible to existing benchmarks. The natural remedy is to employ search tools, enabling agentic visual generation. However, we find that naive search fails: it retrieves indiscriminately, injecting noise into prompts the generator already handles. We trace the root cause to a generator-specific, evolving knowledge boundary: the divide between what a generator can internalize through training and what must remain in external context. Although this boundary is hard to specify in advance, we show that it is discoverable through a teach-then-search co-training framework. Even a minimal version of this co-training recipe produces monotonic improvement, laying the foundation for recursive self-improvement in visual generation that can meet world-knowledge-grounded requests. We release the full dataset, co-training corpus, and search corpus as a replayable harness for tool-augmented, world-knowledge-grounded visual generation.