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

The paper discusses a research project aimed at improving the capabilities of visual generators by addressing their limitation in fabricating what they don't know about the ever-evolving and vast visual world. The authors present SearchGen-20K and SearchGen-Bench, datasets with various failure categories and domains, along with a pre-executed multimodal SearchGen-Corpus-1M for offline research. They find that naive search fails to enhance the performance of visual generators due to a generator-specific knowledge boundary. To overcome this, they propose a teach-then-search co-training framework that can potentially lead to recursive self-improvement in visual generation.

What the Document Says

• Visual generators excel at rendering but often fabricate what they don't know.
• The authors present datasets named SearchGen-20K and SearchGen-Bench, along with a pre-executed multimodal SearchGen-Corpus-1M for offline research.
• On SearchGen-Bench, frontier open generators score poorly due to their inability to handle unbounded, evolving user requests.
• The authors suggest that employing search tools could potentially enable agentic visual generation, but they find that naive search fails as it injects noise into prompts the generator already handles.
• The root cause is traced to a generator-specific, evolving knowledge boundary between what a generator can internalize through training and what must remain in external context.
• The authors propose a teach-then-search co-training framework that can potentially lead to recursive self-improvement in visual generation.
• The authors release the full dataset, co-training corpus, and search corpus as a replayable harness for tool-augmented, world-knowledge-grounded visual generation.

Key Concepts Mentioned

• visual generators
• knowledge bottleneck
• SearchGen-20K
• SearchGen-Bench
• SearchGen-Corpus-1M
• teach-then-search co-training framework

Claims Made by the Document

• Frontier open generators perform poorly on evolving user requests.
• Naive search fails to enhance the performance of visual generators due to a generator-specific knowledge boundary.
• The proposed teach-then-search co-training framework can potentially lead to recursive self-improvement in visual generation.

Evidence Used

• Visual generators excel at rendering but they confidently fabricate what they do not know
• On SearchGen-Bench, frontier open generators score only 21 to 28 out of 100
• Even a minimal version of this co-training recipe produces monotonic improvement

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.