Anyone who has worked a daytime match knows the problem. Half the pitch sits in bright sunlight, the other half in the shadow of the stand, and the camera has to pick one to expose for. Either the sunlit players glare or the shaded ones turn into silhouettes. We cover the wider problem inshadows in live sports; this page is about one specific answer to it, usually searched for as “AI shadow removal”, and about being straight on what it can and can't do.
What “shadow removal” means on a live feed: lifting detail, not repainting
The phrase suggests something dramatic: the shadow disappears and the pitch looks evenly lit, as if the sun had moved. That is not what LivePictureAI does, and we would be wary of any live system that claimed to.
What the AI does is balance bright sunlight and deep shadows in real time. It works frame by frame and treats the sunlit and shaded regions of the same frame differently, lifting detail in the shade and holding back the bright areas, so a player crossing the shadow line stays readable on both sides of it. The shadow is still there. You can still see that one part of the pitch is in shade. What changes is that the shaded part stops being a dark mass and starts showing shirts, numbers and faces again.
That distinction matters for a broadcast. Repainting a scene (inventing an even light that was never there) would mean generating pixels, and generated pixels in a live sports picture are a credibility problem. Lifting and rebalancing works with the signal the camera actually captured. Nothing is added to the scene that was not already in it.
What it can and can't recover
Modern broadcast cameras usually capture more in the shadows than ends up visible on screen. That detail is sitting low in the signal, compressed into the bottom of the range so that the sunlit part of the frame does not blow out. This is where real-time enhancement earns its keep: if the information is in the signal, it can be brought back up to where a viewer can see it, region by region, without dragging the rest of the frame with it.
There are hard limits, and they are worth stating plainly:
- Clipped highlights cannot be recovered. If a white shirt in full sun has hit the top of the signal and flattened into pure white, the folds and the sponsor logo are gone. No processing after the camera can bring back what was never recorded.
- Crushed blacks cannot be recovered either. If the shade has been pushed down to a flat floor with no variation in it, there is nothing to lift. Brightening it only gives you a lighter flat area, often with more visible noise.
- It cannot invent detail. It will not sharpen a player who is out of focus, put back information lost to heavy compression upstream, or make a badly exposed camera look well exposed.
In practice that means camera exposure still matters. The best results come when the cameras are set up to keep some information in both ends of the range, and the AI is left to rebalance it. The processing widens what you can show from a well-captured signal; it does not rescue a signal that has already thrown the information away.
How to evaluate it on your own footage
Polished demo clips are chosen to look good. The only test that really tells you anything is your own footage, from your venues, at the times of day that cause you trouble. When you look at a result, a few questions are more useful than “does it look better?”:
- Can you follow the play across the shadow line? Watch players run from sun into shade and back.
- Does it stay consistent over time? Look for pumping or flicker as the shadow moves or the camera pans.
- Does it still look like your broadcast? Skin tones, grass and kit colours should look natural, not graded.
- What happens in the hard cases? Check the clipped highlights and the deepest shade, and see that it does not pretend to recover what is not there.
- Does it look AI-generated? If anything looks painted on, that is a failure, however striking it is.
We have tested the approach on football footage, and we would rather show you your own. You cancompare original and enhanced frames and send us a clip, and we will return a sample enhancement so you can judge it against the questions above.
Where it sits relative to the rest of the picture chain
LivePictureAI is a processing layer with two setups. In large productions it sits on each camera feed before the vision mixer, so the director cuts with enhanced pictures; in smaller productions it sits on the programme output after the vision mixer. It works with SMPTE 2110, NDI, SDI, RTSP and file-based streams, and it does not require new cameras. It adds at most 100 ms of latency.
It is just as important to say what it does not take over:
- Vision engineers still own camera matching. Iris, colour and matching between cameras stay where they are. The enhancement works on what the cameras deliver; it does not replace the people shading them.
- Exposure decisions still happen at the camera. As above, the processing works best when there is information to work with.
- It is not a grade. It does not impose a look on the programme. It addresses one problem (sunlight and shadow in the same frame) and leaves the rest of the chain alone.
We are validating the system with selected partners following IBC 2026. If your productions fight the same sun-and-shadow problem, that is the conversation we want to have.
Not an effect. Not a filter.
It is easy to file AI video processing under effects: something applied on top of the picture that changes how it looks. That is the opposite of the goal here. It is not an effect, it is not a filter, and it should not look AI-generated. If a viewer notices the processing, it has gone too far.
The honest description is less exciting than “shadow removal” and more useful: real-time AI that rebalances sunlit and shaded regions of the same frame so the detail your cameras already captured is visible to the viewer. Just a better live picture. And the producer decides how much shadow stays.