On the jogging shot, the result is impressively close to finished. The woman and both dogs are isolated well, and the mask stays stable across the sequence. The main weakness is around the hair, where there is a bit of edge fringing, but outside of that, the cutout holds together very nicely.
That is an important point. No automated background removal is perfect on every frame, especially with fine detail. But if the only thing that really needs extra attention is hair refinement, that is already a very strong starting point.
Photoshop vs DaVinci Resolve Magic Mask v2
To see how this compares to Resolve’s newer AI tools, the same jogging clip was also tested with Magic Mask v2. The setup is familiar: go to the Color page, choose the AI mask tool, and click over the subject to tell Resolve what should be included.
In this case, multiple dots were placed over both the woman and the dog so the tool would try to understand the entire subject group. Quality was switched from the faster mode to Better for a cleaner result before tracking forward and backward.
At first glance, it looked promising. But once the tracking started, the mask began to break down. The woman was picked up reasonably well, but one of the dogs started disappearing and parts of the selection flickered or glitched as the shot progressed.
Could that be fixed? Yes, with enough manual correction. But that defeats the whole point of using a supposedly quick AI roto tool on a shot like this.
Why Photoshop won on the first test
On the jogging shot, Photoshop had a clear advantage in two areas:
- Subject completeness, because it kept the woman and both dogs together more reliably
- Mask stability, because the cutout did not wobble or fall apart nearly as much over time
Magic Mask was not unusable, but it was less dependable. The more the subject arrangement changed, the more likely pieces of the mask were to drop out.
A tougher test: bike spokes and a more complex scene
The second example is where things get really interesting. This shot is much harder because it includes a bicycle wheel with thin spokes, overlapping shapes, and multiple people interacting in the frame.
That kind of image usually exposes weak masking instantly. Fine details vanish, holes get filled in, and the separation between foreground elements tends to collapse.
Photoshop handled it far better than expected. The standout detail was the wheel. It managed to preserve the open spaces between the spokes, which is exactly the sort of thing that tells you whether a tool is genuinely doing strong subject extraction or just making a rough silhouette.
How Magic Mask handled the complex bike shot
The same test was repeated with Magic Mask. A lot more input was needed this time, including extra dots placed carefully around the bike so Resolve would attempt to include it. Quality was again set to the better mode before tracking.
But once the track started, the selection began to fall apart. The tool could identify the people to a point, yet it struggled to hold onto the bike and the fine details. The mask glitched, broke apart, and did not come close to matching the cleaner Photoshop result.
That is really the pattern across both examples. Magic Mask can be fast for simpler isolation work, but on complex shapes and multi-part subjects, Photoshop’s cloud-based cutout was more dependable.
