跳到主要内容Skip to main content

怎么判断一张图是不是 AI 生成的How to Tell Whether an Image Is AI-Generated

微型指南 · 约 7 分钟Guide · about 7 min · 引用核验 2026-11-17(仅引用链接与日期) · refs checked 2026-11-17 (reference links and dates only)

单看画面已经不可靠;能站住脚的判断来自「来源、原件、凭证、时间线」四条证据链。Looking at pixels is no longer enough. Defensible judgments rest on four evidence chains: source, original file, credentials and timeline.

怎么做How to

先看来源与凭证:C2PA / Content Credentials 能记录拍摄与编辑;没有凭证不等于假,有凭证也要验签。

查反向图搜与发布时间:AI 图常常找不到更早的出处。

看画面线索:文字、手指、重复纹理、镜面反射、阴影方向;但模型进步会不断让这些线索失效。

看噪声与压缩一致性:拼接图常有不同噪声、重复块或边缘不一致。

检测模型只作辅助:它们对压缩、二次编辑和分布外图片很脆弱,概率不是判决。

最可靠的是原件、拍摄链路与证人;平台应保留上传原件和处理日志。

结论分三档:已证实伪造 / 高度可疑 / 无法判断;不要被迫二选一。

Check source and credentials: C2PA / Content Credentials can record capture and edits; no credential does not prove fake, and a credential still needs signature verification.

Run a reverse image search and check publication dates; AI images often have no earlier origin.

Look at visual cues: text, fingers, repeated textures, mirrored reflections and shadow direction — but model progress keeps invalidating them.

Check noise and compression consistency: composites often show mismatched noise, repeated blocks or edge discontinuities.

Use detectors only as aids: they are fragile under compression, re-editing and out-of-distribution images; a probability is not a verdict.

The strongest evidence is the original file, capture chain and witnesses. Platforms should retain the uploaded original and processing logs.

Conclude in three grades: proven fake, highly suspicious, cannot determine. Do not force a binary answer.

例子与判断Example and interpretation

以下是教学情境,不是本站运行或实测记录。A teaching scenario, not a measured or executed site test.

拿到一张图,先记录「来源、原件、凭证、时间线」四项;四项都缺时,结论只能写「无法判断」,不能写「AI 生成」。For any image, record four things first: source, original file, credentials and timeline. If all four are missing, the conclusion is “cannot determine”, not “AI-generated”.

做完例子后怎么检查How to check your practice

先用自己的话解释上面的判断,再检查输入条件,最后回原文验证。说不清原因时回概念;缺日期、字段或权限时补输入;结论无出处时先撤回结论。Explain the example in your own words, inspect input constraints, then verify against the source. Revisit concepts if you cannot explain why; supply missing dates, fields or permissions; withdraw unsupported conclusions.

按这篇主题,先从“结果缺验证”练习查起For this topic, start with unchecked output · 查看全部六个情境与理由See all six scenarios and their reasoning

出处Sources
  • C2PA · 2.1 规范C2PA 2.1 specification ↗
  • Content Authenticity InitiativeContent Authenticity Initiative ↗
围绕这些概念继续读More on these concepts

怎么给 AI 内容加来源标识How to Label AI-Generated Content