Wildlife AI
From footage to wildlife data: photos, video, and how our AI reads them
Every photo and video your cameras send to the Untamed Data Platform is analyzed automatically: species identified, animals counted, results ready to review. But photos and video aren't the same job for the AI, and knowing when to use each gets you cleaner data and faster results. Here's how it works.
Timing
How long does analysis take?
Photos are analyzed within seconds of arriving. A single image is one look at the scene, so the AI detects the animals, identifies the species, and files the result almost immediately.
Video takes longer. Instead of one frame, we sample many frames across the whole clip for a clearer read on the species. That thoroughness costs time: a clip can take a few minutes, depending on its length. Per-animal tracking — so a group walking past is counted individually rather than missed — is in development.
Either way, it all runs automatically in the background on our European servers. You never wait on a screen — results appear as soon as they're ready.
Photos or video — which is better?
Neither is universally better; it depends on what you want from the data.
- Species identification — video usually wins. More frames mean more chances at a clear, well-framed shot, and we keep the best one. A single photo at an awkward angle gives the AI less to work with.
- Counting individuals — photos (especially bursts of 3–5) are cleaner. A photo is simply "who is in the frame right now," so the count is unambiguous. Counting from video means tracking animals across frames — powerful, but inherently trickier, and something we are still building.
- Behavior and activity — video wins. Movement, direction, and interactions only show up over time.
- Speed, cost, and storage — photos win comfortably. One image is one quick analysis and a small file; a video is many frames, more minutes, and a much larger file — and it drains batteries and SD cards faster.
Our recommendation
For day-to-day monitoring and population counts, photos — ideally short bursts — are the workhorse. They're fast, affordable, give clean per-trigger counts, and are more than enough for reliable species identification. It's also what most camera-trap researchers choose.
Reach for video when behavior or moving groups matter — surveys, anti-poaching, herd dynamics. That's where the extra detail earns its keep, at the cost of slower processing.
One note on counting groups: per-animal tracking — in development — will make video counts far more accurate than a naive frame-by-frame approach, which tends to under- or over-count. Until it lands, and because tracking is harder than reading a single frame, photo bursts stay the most reliable choice when your priority is precise abundance numbers.
