朱雀AI检测

朱雀AI检测 is a browser-based content authenticity checker from Tencent that estimates the likelihood a piece of text or an image was produced by an AI system rather than a person. It targets the growing problem of blurred authorship, ghostwritten academic work, and fabricated visuals circulating through social and news channels.

What it does

Users paste text, upload a document, or drop an image into a clean web interface. The service runs the material through detection models and returns a probability score, highlights passages or regions it consIDErs suspicious, and summarizes the result in a readable report. Tencent describes the engine as combining several algorithms trained on a large corpus of human-written and machine-generated Samples, with periodic updates as new generative models APPear.

Key capabilities

• text detection: Accepts pasted content or txt and docx files, then flags SECtions likely written by assistants such as GPT-series, claude, DeepSeek, Gemini, and Chinese models. Reported text accuracy is in the low-to-mid 90s percent range, though real-world results may vary.
• Image detection: Supports png, jpg, and webp uploads and examines pixel and texture patterns associated with diffusion and GAN generators, plus signs of face-swap manipulation. Claimed image accuracy is above 98%, and results usually return within a minute.
• Risk labeling: OuTPUts percentage scores and tiers such as high, medium, or low risk, with highlighted excerpts or marked areas on the image.
• Batch and audit use: designed so instructors, editors, and data teams can screen many submissions or filter training datasets.
• privacy stance: Tencent states submitted material is used only for the live check and is not retained for model training, though independent verification is adVisable.

Who it is for

The tool suits educators checking assignments and theses, newsroom editors verifying contributed text and photos, publishers and agencies confirming originality, legal and medical teams screening documents or scans, and AI teams Cleaning training data so synthetic samples do not degrade model output. Casual users curious about whether an article or picture is machine-made can also try it.

How to get started

1. Open the official Tencent Matrix page in a desktop browser.
2. Choose the text or image tab and paste, dRAG, or upload your file.
3. Review the probability score, highlighted passages, and risk notes.
4. For high-stakes decisions, confirm findings manually or with a second tool.

Pricing and access caveat

朱雀AI检测 is free to use and offers a daily allowance of checks for both text and images, reportedly around twenty text and thirty image scans per day, with no CLIent installation needed. Quotas, accuracy figures, and language support may change without notice, and the interface is primarily Chinese, so non-Chinese speakers may need translation help. Detection is probabilistic: short human writing, stylized illustrations, heavily edited photos, and low-resolution files can produce false positives or negatives.