AI detector · Deepfake · Alibaba Cloud

Deepfake Face

Upload a portrait to screen for AI face-swap and synthetic-edit risk. You get per-face labels and confidence scores—useful for content review, trust and safety, and fraud awareness (not a legal determination).

Your upload

Clear, front-facing photos work best. One run analyzes the whole frame; multiple faces each receive a row on the result page.

Sample preview

Illustrative asset for the tool—your own uploads are analyzed on the left when you run detection.

Deepfake face detection sample preview

How it works

Powered by Alibaba Cloud DeepfakeFace (facebody): low-level texture cues and higher-level semantics are combined to estimate whether each face behaves like a real photo or like swap / generative editing. Default label threshold is around confidence 50—scores below lean authentic; above lean synthetic.

01

Upload a clear image

PNG, JPG, JPEG, BMP, GIF, or WEBP up to about 10 MB. Prefer resolution above 256×256 px with a face region larger than ~128×128 for stable scores.

02

Cloud analysis

The image is sent securely with your account token. The model returns face count, per-face label, confidence (0–100), optional boxes, and message tips for quality issues.

03

Review results

Open the result page to read each face. Treat output as probabilistic—combine with context, policy, and human review for serious decisions.

FAQ

Labels, confidence, multi-face behavior, credits, file limits, and responsible use.

Screen the next image

Use Deepfake Face as one signal in your review workflow—then iterate with clearer photos if tips flag blur, occlusion, or angle issues.

Back to upload