AI Image Detector
Model:
Upload an image and our vision model checks it for signs of AI generation — warped hands, garbled text, impossible reflections, diffusion artifacts. Get a probability score and the specific signals it spotted.
—
——
Signals detected:
This is an AI-powered estimate, not a court-admissible forensic test. False positives and false negatives happen — use judgment alongside it.
Advanced options
結果
Tokens running low.
Get More Tokens
もっと欲しい？
Sign up free for 10,000 tokens
無料登録
ご要望を処理中...
Paste or upload any image and find out whether it was AI-generated. Free, no account, instant result with confidence score.
使い方 AI Image Detector
1
入力を入力
テキストを入力、ファイルをアップロード、または必要なことを記述します。アカウントは必要ありません。
2
クリックして生成
私たちのAIは、最良のオープンソースモデルを使って、あなたの要求を数秒で処理します。
3
ダウンロードと共有
結果をダウンロード、コピー、共有できます。個人的、商業的な使用は無料です。
Use this tool via API
Automate this tool from your own code. OpenAI-compatible REST endpoint, Bearer-token auth, no extra SDK required. Token costs match the web interface.
curl -X POST https://api.free.ai/v1/image/generate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"prompt": "A sunset over mountains", "model": "sdxl"}'AI Image Detector — FAQ
Upload any image and our vision model checks it for signs of AI generation — warped hands, garbled text, impossible shadows, diffusion artifacts. You get a probability score, a verdict, and the specific signals it spotted.
Yes. Each check costs about 500 tokens. Signed-up users get 10,000 tokens/day free.
Good for catching obvious AI imagery from older models (Stable Diffusion 1.x, early Midjourney, etc.). Much harder on state-of-the-art models like FLUX and Nano Banana — those can pass the test. Treat it as a heuristic, not forensic proof.
Hand and finger anatomy, teeth, symmetric eyes, text legibility, impossible reflections and shadows, overly smooth skin, warped backgrounds, and repeated textures — classic tells of diffusion model outputs.
0-40% = likely real. 40-70% = uncertain, check manually. 70-100% = likely AI-generated. The confidence badge tells you how sure the model is about that estimate.
Modern AI images are hard to distinguish. If the model isn't confident either way, it says so rather than guessing. That's honest — other detectors will give you false confidence.
No. This is a quick heuristic, not a forensic tool. Courts require provenance-based verification (C2PA, cryptographic signing). Use this to sanity-check, not to settle disputes.
Qwen 2.5-VL — a multimodal vision-language model. We prompt it specifically to look for the telltale signs of generative image artifacts.
JPG, PNG, and WebP up to 10MB. Higher-resolution images give the model more detail to analyze.
Occasionally — stylized portraits, heavily edited product shots, and low-quality JPEGs can trigger false positives. Always read the signal list before concluding.
No. Uploads are analyzed and discarded immediately. Only the result is kept (if you sign in, in your generation history).
Yes — POST an image to /v1/image/describe/ with mode=ai-detect. See /api/ for docs.
このツールをどう評価しますか？