← Analysis types · Start · One post
Pictures and video are not decoration. They are read the same way captions are: who is in them, what they say, and where they are. A post can mention someone in the caption, show their face in the photo, and tag a place — we keep all three.
We save the file first, then analyse it. Saving and analysing are separate: media can sit in storage even if analysis fails, and we can re-run analysis without downloading it again.
flowchart LR save["Save the file"] --> still["Look at the picture / frames"] still --> faces["Faces"] still --> text["Text in the picture"] still --> scene["What's in the scene"] still --> place["Places"] faces --> match["Match to author photos"]
Images
| What we look for | Meaning | What you can do with it |
|---|---|---|
| What’s in the scene | Objects, products, logos, crowds, protests, meals, screenshots… | Find posts by what is shown, not only by the caption. |
| Face presence | Whether people appear, roughly how many, whether faces are clear or in a crowd. | Filter to posts that show people vs posts that show only products or landscapes. |
| Who it looks like | Faces in the post are compared to the author’s profile photo (and other known faces for that account). | See if the account owner is in the post, or if the photo is of someone else. Helps tell original posts from reshared images. |
| Text in the image | Words on packaging, signs, screenshots, memes, stickers, overlays, watermarks. | Search those words with the caption. Catches @handles and slogans that never appear in the post text. |
| Mentions in the picture | Handles, names, and logos readable on-screen, plus people the post tags. | Tied to the same people and brands as caption mentions. |
| Place tagging | Places the platform tagged, plus places we can see (a landmark, a shop front, a city skyline) or read (a street sign). | Map and filter by location even when the caption never names it. |
| Safety | Whether the image looks unsafe. | Held for review. Does not delete the post. |
No image means this step is skipped.
Video
Video gets the image passes on sampled frames, plus what was said.
| What we look for | Meaning | What you can do with it |
|---|---|---|
| Transcript | Speech turned into text. | Search spoken words, not just the caption. |
| On-screen text | Captions burned into the video, lower-thirds, signs, phone screens. | Folded into search with the transcript. |
| Faces in frame | Whether people appear on camera, and whether a face matches the author’s profile photo. | Same owner-vs-other split as photos. |
| Spoken mentions | Names, brands, and handles said out loud. | Counted with caption mentions. |
| Places | Tagged location, plus places named in speech or visible in frames. | Same place filters as images. |
Mentions, faces, and authors
A “mention” is anyone the conversation is about — in text, on screen, or in the room.
| Kind | Example |
|---|---|
| Written | The caption tags @alice or names a brand. |
| On-screen text | A screenshot of a chat, a protest sign, a labelled product. |
| Face | A person appears in the photo or on camera. If that face matches the author’s profile image, we treat it as the author appearing in their own content. If it does not, it is someone else in their media. |
| Place | The post is tagged at a café, or the video is clearly shot in a known square. |
Profile photos collected with the author are the reference for face matching. We do not use private photos — only the public profile image and faces already in that account’s public posts.
See Analysis types for how this sits next to sentiment and stories, and One post for when it runs on a single item.