[Now in Beta] Introducing Smooth Mode in Object Mask

Hi, I've done some testing on high res media. There is a big issue with the current model or it's implementation that causes severe aliasing as soon as ‘tracking’ starts. I’ve made a post about it in the AE Beta board too as it concerns both packages.

But here's the image for quick reference.

This should really be fixed before it can be deemed production ready.

Where is my Mask Quality my lovely Adobe team

Hi Adobe Team,

I’d like to suggest an enhancement to Photoshop’s masking workflow that could benefit both intermediate and advanced users. Currently, creating precise masks from colour data often requires multiple steps — using channels, Apply Image, Selective Color, and Levels — which can be fiddly and unintuitive.

My idea is an Advanced Masking Panel where users could:

• Select a source channel (RGB, Lab, or individual channels)

• Adjust hue range sliders to isolate specific colours

• Adjust luminance sliders to target highlights, midtones, or shadows

• Preview the mask live and refine it interactively before applying

This would allow users to isolate very specific pixels — for example, only the brightest blue highlights in wave spray — without inadvertently selecting background elements. The workflow would remain non-destructive, visual, and much more accessible, reducing trial-and-error while retaining full creative control.

It’s a small enhancement conceptually, but it could dramatically simplify advanced masking, making Photoshop more intuitive and powerful for a broad spectrum of users.

Thanks for considering!

Thanks a lot for adding the ‘smooth’ model which for many users is probably the most important. I had a play with it briefly and I’m really satisfied by it’s stability and ability to resolve edge detail, motion blur and going in and out of focus.

I do however have a few things to note.

Workflow

The current limitation in switching models is something developers should address. I understand that the modes require different AI models but the part where the user decides which regions are objects to track is data that should be stored for each object mask. When switching modes, these definitions should automatically be reapplied for the new model so the user only needs to re-track, not redefine. Take a look at how this is tackled in DaVinci Resolve. There a clear UI is present with which frames have sample points from which the model would start detecting and tracking and you can jump to reference frame to check mask quality before tracking.

Minor Issue

An interesting observation was that on a particular shot, I tracked a talent but another talent came into frame which the model started to include even though they were fairly spread out. On one single frame there is also a visual glitch where it looks like the model has a certain search area/bounding box and on that frame it probably detected another person and thought to include it.

Workflow wise this would obviously come down to adding a negative input and track from there instead, but if the model itself could also be fine-tuned to understand that if it tracks a person it should only keep tracking a single person unless the user adds more.

If you're interested in looking at the shot, it's clip number 09 from here.

I don’t like the idea that you’ll have to reapply and retrack the mask if you choose smooth after sharp was tracked. Why couldn’t you just add a reduce chatter and smooth sliders to enhance the mask non-destructively? This way you always have to choose which sometimes may ruin your workflow. A better thing would have been to be able to keyframe the reduce chatter and smoothify parameters in a mask for situations when you need to gradually go from sharp to soft and back, loke when an object goes out of focus. Feather won’t always help because it changes the edge of the mask, not the details