media-verification

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Media Verification

媒体验证

Confirm that a piece of media is real, original, and shows what it claims. Standard practice for journalism and disinformation work.

确认某一媒体内容真实、原始且与所声称的内容相符。这是新闻工作和虚假信息治理中的标准流程。

The checklist

核查清单

Provenance — is this the original or a repost? Run
```
reverse-image-search
```
(video: search keyframes) to find the earliest copy and its true date and caption. Recycled old media in a new context is the most common deception.
Metadata — run
```
exif-metadata-analysis
```
for capture time, GPS, and editing-software traces (often stripped by platforms — absence isn't proof).
Location — does the scene match the claimed place? Verify with
```
chronolocation
```
.
Time — do shadows, weather, and season fit the claimed date? Cross-check weather archives.
Internal consistency — signage language, currency, plates, and license details should all agree with the claimed setting.

来源溯源——这是原始内容还是转发内容？运行
```
reverse-image-search
```
（视频：搜索关键帧）以找到最早的版本及其真实日期和说明。在新语境中重复使用旧媒体是最常见的欺骗手段。
元数据——运行
```
exif-metadata-analysis
```
获取拍摄时间、GPS信息和编辑软件痕迹（平台通常会剥离这些信息——缺失不代表没有问题）。
地点验证——场景是否与声称的地点相符？使用
```
chronolocation
```
进行验证。
时间验证——阴影、天气和季节是否与声称的日期相符？交叉核对气象档案。
内部一致性——标识语言、货币、车牌和许可细节都应与声称的场景一致。

Manipulation & AI

篡改与AI生成内容

Edits — error-level analysis (FotoForensics), clone/splice artifacts, inconsistent lighting/shadows and reflections, warped edges near pasted objects.
AI-generated — check hands, teeth, ears, text/logos, jewelry, and background coherence; watch for too-smooth skin and nonsensical fine print. Detection tools help but are not reliable alone — corroborate.
Video/deepfake — unnatural blinking, mismatched lip-sync, flickering face edges, inconsistent head/neck lighting; inspect frame by frame.

编辑篡改——错误级别分析（FotoForensics）、克隆/拼接痕迹、不一致的光线/阴影和反射、粘贴物体边缘的变形。
AI生成内容——检查手部、牙齿、耳朵、文字/标志、珠宝和背景连贯性；留意过于光滑的皮肤和无意义的小字。检测工具虽有帮助，但不能单独作为可靠依据——需多方佐证。
视频/深度伪造——不自然的眨眼、唇形不符、面部边缘闪烁、头部/颈部光线不一致；逐帧检查。

Verdict

结论

Report a graded conclusion — authentic / altered / misattributed / AI-generated / inconclusive — with the specific evidence, never a bare yes/no. When unsure, say inconclusive; a wrong "verified" is worse than an honest gap.

给出分级结论——真实/篡改/错误归因/AI生成/无法确定——并附上具体证据，切勿仅给出是或否的简单答案。若不确定，应说明无法确定；错误的“已验证”比坦诚的不确定性更糟糕。