How to detect DALL-E images

DALL-E 3 (integrated into ChatGPT) and DALL-E 2 have distinct visual signatures and always embed C2PA content credentials, making them among the easiest AI generators to confirm definitively.

C2PA content credentials — the definitive tell

Every image generated by DALL-E 3 through ChatGPT carries a C2PA "Content Credentials" provenance block signed by OpenAI. This is not optional — it is embedded automatically. When you upload such an image to AiPicDetect, the metadata panel will show a C2PA block. The content credentials include:

• The generator name (OpenAI / DALL-E)
• A creation timestamp
• A cryptographic signature from OpenAI's root certificate

A C2PA block from OpenAI is the strongest possible machine-readable proof of AI origin. No other signal is needed.

Important caveat: if the image has been screenshotted, re-saved through certain tools, or heavily edited, the C2PA block is stripped. The absence of a C2PA block does not prove the image is real — it may just mean the block was removed. Use the pixel-level detector score alongside metadata.

Visual characteristics

DALL-E 3 images tend to be:

• Brighter and more saturated than Midjourney — DALL-E defaults to a high-contrast "editorial illustration" aesthetic.
• Better at text: DALL-E 3 handles short in-image text more reliably than earlier generators. Signs and labels may look plausible.
• Flatter depth: DALL-E 3 portraits often have less convincing bokeh and background separation than Midjourney v6.
• Prompt-literal: DALL-E 3 follows prompts more literally, which can produce compositions that feel rigidly posed.

DALL-E 2 images

DALL-E 2 pre-dates C2PA and does not embed credentials. Visually it is easier to detect: lower overall resolution, characteristic smooth skin, flat lighting, and frequent hand/face distortions. AiPicDetect's detector scores DALL-E 2 outputs as High-confidence AI reliably.

Quick reference

Check a DALL-E image →

Cập nhật lần cuối 2026-09-13 · mã nguồn trên GitHub

Dùng thử công cụ phát hiện