Unverified50% confidenceEventExact time
有用户使用Claude和Qwen自家的语言模型测试,均能识别出Qwen-Image-2.1图像中的纹理缺陷
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/20/2026
First Seen
Valid until: 12/19/2026
Sources
Related Entities
Related Claims
Unverified一位研究者使用Qwen-Image-Edit和HiDream O1两款生成模型制作合成图像进行深度伪造检测器基准测试70% similarUnverifiedQwen-Image-2.1生成的图像存在明显的点状(dotted)和抖动(dithering)纹理缺陷68% similarUnverifiedQwen系列模型和Llama-3.2-3B都能通过该方法暴露出人类可感知的风格轴65% similarUnverifiedQwen模型在空间理解方面存在一贯的弱点,在3D房间渲染测试中出现场景翻转和物体位置错误64% similarUnverifiedQwen-Image-3.0能精细刻画人物照片中的毛孔、发丝等微观特征63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937461API
curl https://kongchang.com/api/v1/knowledge/claims/937461MCP
get_claim(id=937461)