Check AI information first
Meta's labeling approach covers Facebook as well as Instagram and Threads, and it distinguishes AI generation from some AI editing when presenting AI information. Review the available label and the poster's own description before assessing a downloaded copy. A label supplies context about production; it does not prove that the caption, event, ownership or sharing permission is correct.
Find the earliest useful source
A widely shared photo may have passed through many accounts before reaching your feed. Look for earlier copies of the same image to find where it first appeared, and compare the claim with trusted reporting or the original publisher's description. Ask the poster for the source link or export if appropriate. Image Evidence does not search for people, identify faces or authenticate Facebook accounts.
Work from the best file you can get
A Facebook post link is not a media file, and images saved from a feed are compressed copies. When the question matters, ask the creator for the original JPG, PNG, WebP or HEIC export, or the original video file. Keep the post reference separately so any result can be reviewed alongside its context.
Expect signal loss and uncertainty
Reposts, screenshots and transcoded copies may change or omit provenance data, and we have not found a Facebook-wide guarantee that C2PA data or SynthID signals survive sharing. Finding a provenance container is different from validating its signature, and a missing container does not prove a camera capture. Image Evidence has not published validated detection performance for Facebook images.
Choose a proportionate next step
For a casual image, the creator's explanation may settle the question. For consequential claims, ask for corroborating records and let the source respond. Do not use an isolated score to deny an opportunity or publicly accuse a person. Image Evidence does not train on user content and provides no watermark removal or detector evasion.
Does Facebook label AI images?
Meta labels some content on Facebook with AI information when people disclose AI use or when it detects industry-standard signals. Not every AI image carries those signals, and for content that was only edited with AI the information may appear in the post menu instead of on the image.
How do I check if a Facebook photo is real?
Separate two questions: was the image generated or edited, and is the post's claim true? For the first, look for AI information, ask for the original file and check any provenance records. For the second, find the earliest source and compare it with independent reporting. A real photo can still be shared with a false caption.
Why does an AI detector give different results for the same Facebook image?
Compression, resizing and screenshots change the pixels that detectors read, and different tools are trained on different data. Small changes can move a borderline score. Consistent results across versions of an image are more meaningful than a single high or low number.