Spot and count animals live from your DJI drone

By Taner Özel · Updated October 21, 2026

DJI Live Bridge can now look at your drone's live picture for you. AI Vision finds goats, sheep, cattle, horses, donkeys, poultry and other animals, draws a labelled box around each one on your stream and the virtual camera, and keeps a count. It runs on your own computer; no picture or result is sent anywhere.

What AI Vision does

When you fly over a pasture, counting the herd or finding the one goat behind a bush is hard on a small controller screen. AI Vision does the looking: it checks the live picture over and over, once or twice a second with OWLv2 and up to ten times a second with the lighter models, and marks each animal it recognises.

• Boxes and labels with the species, a number for each animal and how sure the model is, for example Sheep #7 78%.
• An animal counter in the corner of the picture: how many are visible now, by species, and how many different animals it has seen since you reset it.
• On every output. The boxes are drawn into the live stream to Instagram, TikTok, YouTube and your other destinations, into the virtual camera and into the preview in the app.
• Boxes are optional. You can hide the boxes or the counter and keep the counting running in the app.

AI Vision only adds a layer on top of the picture. The drone video itself is never held back to wait for the analysis.

Which animals it recognises

Most ready-made detection models are trained on the same public photo collections, and those have no goat at all. A goat comes out as a sheep, or as nothing. That is why the default model on a Mac is OWLv2, a model that is told by name which animals to look for.

You can switch models under Detection settings → Model; the list shows how many species each one knows. For mixed herds of goats and sheep, use OWLv2. For cattle and horses only, D-FINE X is about five times faster. Under Detected species you can limit the search to the animals you care about.

Wild savanna animals such as giraffes, zebras and elephants are not in the list on purpose: over a herd of cattle they only produced false alarms.

How to turn on animal detection

1. Connect your droneSend the live picture from DJI Fly to the app over RTMP as usual; the address is on the Go live tab. The details are in the DJI Fly RTMP guide.
2. Switch on AI VisionOn the Go live tab, find the AI Vision · Animal Detection card and turn on Animal detection.
3. Let it download and prepare the modelThe first time, the model is downloaded and prepared for your computer. With OWLv2 this can take several minutes; the card shows the progress. Later starts take about a minute.
4. Set the thresholdDrag Confidence threshold down to about 0.25–0.35 for shots from the air, where animals are small. Raise it if you see boxes on bushes or rocks.
5. Go liveStart the broadcast or the virtual camera. If you were already live, the boxes join the stream the next time you start it; the virtual camera shows them right away.

No drone nearby? Try with a video file on the Go live tab plays a recorded clip through the same path, so you can try AI Vision at your desk.

Flying for a good count

• Height matters most. From very high up an animal is only a few pixels. The app still finds it, but whether it is a goat or a calf becomes a guess. Fly lower, or zoom in, when the species matters.
• Fly slowly. The default model looks about once or twice a second. A slow pass lets it follow each animal instead of losing and re-finding it.
• Reset before you count. Press Reset count when you start a pass over the herd. An animal that leaves the picture for a few seconds and comes back may be counted again, so treat the unique count as a good estimate, not a census.
• Use the full screen. Double-click the preview, or press the button in its corner, to see the picture and the boxes across the whole screen. Press Esc to go back.

Privacy and requirements

• On your computer only. Detection runs locally. No picture, detection or count is uploaded. The only network use is the one-time model download.
• Mac: the models run on the Mac's graphics and neural hardware. An Apple Silicon Mac is recommended. OWLv2 needs about 4 GB of free disk space once prepared.
• Windows: the lighter YOLOX models run on the processor. OWLv2 and the other large models are offered only on a Mac, so goats, donkeys and poultry are not recognised on Windows yet.
• Off means off. While AI Vision is switched off, no model is loaded and your stream is exactly as before.

Questions

More guides