Unverified50% confidenceTradeoffExact time
In text-to-image generation, excessively pursuing reward scores causes generated results to trend toward homogeneity, while single-mindedly pursuing diversity may produce a large number of low-quality images.
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
Unverified该方法依赖高质量可微的人脸相似度模型作为奖励来源，单步计算成本较高，对画风、构图等奖励信号不明确的属性未必适用73% similarUnverified低质量或过度压缩的训练图像会导致模型在细节生成上产生伪影、色彩失真乃至风格理解偏差72% similarUnverifiedWhen image quality reaches the point where the human eye cannot distinguish real from fake, the marginal perceived benefit to users from improvements in sharpness and realism diminishes.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/975676API
curl https://kongchang.com/api/v1/knowledge/claims/975676MCP
get_claim(id=975676)