izdeu.ai · video and photo archive search
Find any fragment in a video or photo archive by describing it
Find the moment you need in any archive — by description, not by tags.
izdeu.ai analyses speech, imagery, on-screen text, faces and voice. You can search by scene, action, object and context.
Bring 10 hours of your video. In a week we'll show you the result on it.
In use in newsrooms, on construction sites, in security teams, in the public sector and for personal archives.
- Data stays in Kazakhstan
- Kazakh on par with Russian
- Accurate to the second
The archive exists, but nothing can be found in it
- The archive grows, and finding anything in it becomes impossible.
- Nobody fills in tags and descriptions by hand.
- Keyword search does not match the content: "storm" does not find "hurricane".
- The right moment is hunted through hours of manual scrubbing.
How it works
The system builds a search layer on top of the archive. The files stay where they are; what changes is how you reach them.
- Loading the archive. Video and photos are connected from storage, cameras or cloud.
- Automatic analysis. The system recognises speech, imagery, on-screen text, faces, objects, actions and voice.
- A data layer for search. All the signals come together in one meaning space, tied to time.
- Search by description. A plain-words query returns the fragment, accurate to the second.
Instead of watching a long video, the user opens the exact episode straight away.
A search for "cat on a roof" finds shots where nobody wrote a word about roofs.
What the system can do
- Search by scene description, action and context.
- Search by objects and people in frame.
- Search by speech: any spoken phrase becomes a key.
- Search by on-screen text: captions, signage, documents.
- Face recognition with grouping by person.
- Voice identification, including off-camera speech.
- Assembling a clip from a text description.
- Automatic tracking of the events you define.
Where it is used
- TV, media and video production. Find a story by description, quote or speaker.
- Construction and development. Pull the moment you need out of site footage: a work stage, a violation, an incident.
- Security and surveillance. Find an incident in camera footage by describing the situation.
- Public sector. Handling citizen reports and city footage in situation centres.
- Personal archives. Search your own photos and video in plain words and put together keepsake films.
Kazakh on par with Russian and English
A Russian query finds Kazakh newsreel; a Kazakh query finds Russian broadcast.
Kazakh speech is transcribed, not skipped. Off-the-shelf systems either stay silent on Kazakh or return nonsense — that was separate work, and it is done.
Every fragment is described in all three languages, so you can search in the one you think in.
The data stays with you
The system is installed inside the client's own perimeter or in a data centre in Kazakhstan.
- Video, photos and faces never leave your network and are not sent to outside services.
- Access is granted by role: who searches, who uploads, who administers.
- Backups and database integrity checks are part of the installation.
The models run locally; search needs no internet connection. That is a requirement for government bodies and security services.
How a rollout goes
- Pilot, 3–6 weeks — 10–15 hours of your video, processing set up, search checked against your own queries.
- First launch — the main archive, access rights, staff training.
- Scaling — the whole collection, integrations, fine-tuning on your own people and objects.
Timelines are confirmed after we assess the volume and formats.
Common questions
- Can a video fragment be found by description?
- Yes. The system parses the content of the video, so you can search by scene, action, object or line of speech. What comes back is a specific moment, not the whole file.
- How do you search a large archive?
- The archive is connected once and processed on its own. After that you search the whole body of material in plain words.
- Is Kazakh supported?
- Yes. Kazakh speech is transcribed and takes part in search on the same footing as Russian and English.
- Can a person be found by voice when they are not on screen?
- Yes. The speaker is recognised by voice, including in off-screen narration.
- Where is the data stored?
- Inside the client's perimeter or in a data centre in Kazakhstan. For media, the public sector and security services this is usually a condition of the deal.
- Does it work for a personal archive?
- Yes. Search over your own photos and video works the same way, and the system will also cut a film for an occasion from a description.
What already works
The system is built and running on a real archive:
- 33 hours of video indexed
- 17,000+ scenes detected
- 13,000+ historical photographs
- a query answered in a second
At the meeting we open the system and type a query on your own subject.
We'll show it on your archive
Send us 10 hours of material and we'll show the result on it. The conversation commits you to nothing.
Request a demo Telegram channel