Magic Mask v2 - Slower and repeatedly losing the mask

I struggled with Magic Mask v1 as occasionally it would go weird with the mask, but v2 seems to be a LOT worse for me.

Tracking takes forever! a 5 minute talking head and I'm looking at 4 hours to properly track.

That's bad enough but then it constantly loses the mas for no apparent reason and suddenly my presenter has completely vanished from the scene.

Yesterday everything looked good until I went to render when it suddenly decided it needed to retrack everything again. That render ran for 9 hours!

Today everything seemed fine. I've been adding some material to the end of the video and I go back to the central part and AGAIN suddenly the presenter has vanished! What is causing this? If I look at the Colour page my Magic Mask node is still there going to my Alpha node but the Tracking line has mysteriously changed from blue to red. Why, when I haven't touched that part of the timeline? I will now have to wait another 4 hours for it to retrack the clip for the fourth time in as many days.

When I try to redraw the mask it won't let me select a point giving the following error message: "ADDING POINT FAILED: Adding a point faile as the reference frame mask has not been regenerated. Either regenerate the object mask or track the mask prior to adding or removing points."

It seems my only option then is to remove the nodes, reset the base and then add Magic Mask again from scratch. It's such a huge time switch. ChatGPT tells me this is a common and oft-reported issue with v20 caused by cache issues after some sort of crash, which is no real help at all. UPDATE: Just retracking works and the original mask suddenly is shown highlighted once the tracking starts.

Is anybody else having this problem? Does anybody know how to fix it?

It's kind of a crap shoot when you don't post your system specs in your signature as requested. Or when you don't mention what footage you're using. Or share a video or a screen shot.

Generally speaking, MM2 is about half the speed of MM1. But it is also MUCH better. Typically you don't track a whole video because that's really resource intensive as is seen here. I would cut the video up into smaller bits and track them all individually.

Losing tracks when you edit a MM clip is how it works. There are workarounds for this that work perfectly.

Rendering in Delivery is not a great idea because, yeah, it is very slow. Get your video ready to export and then export. You can render in place or cache your mask to speed delivery up greatly.

Are you adding the point on the reference frame? The error is saying to do that but you don't mention whether or not you did.

ChatGPT is not the arbiter of software. It's famous for just making things up to sound like it knows what it's doing. Kinda like a particular orange turd we're dealing with over here. I'm trying to use that and Gemini to do a Fusion 360 (the CAD program) project and they're throwing out bad instructions left and right. "Oh, you're right. Golly I am sorry I missed that one"... ten times in a row.

TLDR: Post system specs. Cut your clips into smaller ones. BONUS: Turn each clip into a compound clip, right click 'Open in timeline', mask, close that timeline and don't open it again. Your mask will save even when you do edits on your main timeline.

Pretty much all Galin said. And ChatGPT is massive red flag. You have refernce manual. Use that next time.

I've been using magic mask all day today and got that error a handful of times. I had to reset the node and try again. Not very impressed with magic mask considering there's AI like Sam2 out there that blows it out of the water and they're not even focused on professional video production. Magic mask has the potential to be the most useful tool in davinci resolves arsenal and now I'm having to seek alternatives.

It's kind of a crap shoot when you don't post your system specs in your signature as requested. Or when you don't mention what footage you're using. Or share a video or a screen shot.

Generally speaking, MM2 is about half the speed of MM1. But it is also MUCH better. Typically you don't track a whole video because that's really resource intensive as is seen here. I would cut the video up into smaller bits and track them all individually.

Losing tracks when you edit a MM clip is how it works. There are workarounds for this that work perfectly.

Rendering in Delivery is not a great idea because, yeah, it is very slow. Get your video ready to export and then export. You can render in place or cache your mask to speed delivery up greatly.

Are you adding the point on the reference frame? The error is saying to do that but you don't mention whether or not you did.

ChatGPT is not the arbiter of software. It's famous for just making things up to sound like it knows what it's doing. Kinda like a particular orange turd we're dealing with over here. I'm trying to use that and Gemini to do a Fusion 360 (the CAD program) project and they're throwing out bad instructions left and right. "Oh, you're right. Golly I am sorry I missed that one"... ten times in a row.

TLDR: Post system specs. Cut your clips into smaller ones. BONUS: Turn each clip into a compound clip, right click 'Open in timeline', mask, close that timeline and don't open it again. Your mask will save even when you do edits on your main timeline.

It's kind of a crap shoot when you don't post your system specs in your signature as requested. Or when you don't mention what footage you're using. Or share a video or a screen shot.

Generally speaking, MM2 is about half the speed of MM1. But it is also MUCH better. Typically you don't track a whole video because that's really resource intensive as is seen here. I would cut the video up into smaller bits and track them all individually.

Losing tracks when you edit a MM clip is how it works. There are workarounds for this that work perfectly.

Rendering in Delivery is not a great idea because, yeah, it is very slow. Get your video ready to export and then export. You can render in place or cache your mask to speed delivery up greatly.

Are you adding the point on the reference frame? The error is saying to do that but you don't mention whether or not you did.

ChatGPT is not the arbiter of software. It's famous for just making things up to sound like it knows what it's doing. Kinda like a particular orange turd we're dealing with over here. I'm trying to use that and Gemini to do a Fusion 360 (the CAD program) project and they're throwing out bad instructions left and right. "Oh, you're right. Golly I am sorry I missed that one"... ten times in a row.

TLDR: Post system specs. Cut your clips into smaller ones. BONUS: Turn each clip into a compound clip, right click 'Open in timeline', mask, close that timeline and don't open it again. Your mask will save even when you do edits on your main timeline.

Gotta love the smart arse replies I keep seeing being posted onhere. A screenshot after the app has completely crashed? Yeah that'll work

I DID use ChatGPT before coming on here. It just told me that yeah, a lot of people have reported the problem and that I should just retrack the whole thing or delete the mask and try again. That takes hours and is precisely why I posted here. ChatGPT has limited access to recent data and this is a new release of DaVinci so it's not the best when it comes to a software release that's just a few days old

I'm happy to post details of my system but given so many other people apparently have the same issue that seemed irrelevant, but heh! go ahead and whinge about it because that will dramatically and immedilatey identify what the problem is (rolls eyes)

One thing I can say is that the rendering might be taking forever because I had to turn off NVidia hardware usage in Settings for "decode H.264/H.265 using hardware acceleration". For years I've had an issue with DaVinci introducing random "Media not available" dropped frames that stop rendering any output - always on clips sourced from Google Studio or YouTube. Finally found a suggestion that turning off this setting (which defaults to using Nvidia) fixes the issue - and it did. I guess not using hardware acceleration could be significantly slowing down the rendering process with Magic Mask, but when the alternative is for it do have issues with random frames that stop you rendering anything at all that isn't an option

Magic mask can be useful, but it is prone to being lost for any number of reasons (related to changing the source video that is fed to the color page). There is a technique where the mask is saved out as separate media and then brought back in an applied, thus bypassing the issue. Then you take full control over when you need to redo the mask. I'll see if I can find the YouTube tutorial and post it below. Yes, it's a workaround but it might just save your sanity.

Add - From Creative Video Tips:

Also, note in the comments that there is an easier way to generate the black and white mask rather than using the color generator but the basic idea is solid.

I should also note that using the luma matte on the color page may not work quite as nicely as Chadwick demonstrates unless the clip and its alpha are both the same length and both start at the beginning of the media (rather than being a subclip of the original media). I haven't tried this in v20 but from memory I recall that the luma matte may not respect the source timecode when aligning on the color page and you then have to use the offset controls for the luma matte to aline the footage with the matte. I'll give it a try and report back when I get a chance.

Update - yep - v20 is no different. If you add the matte file as a matte for the original source (this is a seldom discussed option available in the media storage panel - it binds the matte to the source file), Resolve ignores timecode and assumes that the matte file and source media start on the same frame (which wouldn't usually be the case with a matte generated in Chadwick's example). As I mentioned, you can change the matte alignment with the controls on the color page OR you can render-in-place the source clip which will then put its first frame in alignment with the first frame of the matte clip. I realize this all sounds like a crazy amount of effort but it really isn't. I find using the rendered matte in Fusion to be the simplest way of using an external matte but Chadwick shows us how flexible Resolve is with 3 different ways of accomplishing this task!

Thanks for this. Useful to know.

Magic mask can be useful, but it is prone to being lost for any number of reasons (related to changing the source video that is fed to the color page). There is a technique where the mask is saved out as separate media and then brought back in an applied, thus bypassing the issue. Then you take full control over when you need to redo the mask. I'll see if I can find the YouTube tutorial and post it below. Yes, it's a workaround but it might just save your sanity.

Add - From Creative Video Tips:

Also, note in the comments that there is an easier way to generate the black and white mask rather than using the color generator but the basic idea is solid.

I should also note that using the luma matte on the color page may not work quite as nicely as Chadwick demonstrates unless the clip and its alpha are both the same length and both start at the beginning of the media (rather than being a subclip of the original media). I haven't tried this in v20 but from memory I recall that the luma matte may not respect the source timecode when aligning on the color page and you then have to use the offset controls for the luma matte to aline the footage with the matte. I'll give it a try and report back when I get a chance.

Update - yep - v20 is no different. If you add the matte file as a matte for the original source (this is a seldom discussed option available in the media storage panel - it binds the matte to the source file), Resolve ignores timecode and assumes that the matte file and source media start on the same frame (which wouldn't usually be the case with a matte generated in Chadwick's example). As I mentioned, you can change the matte alignment with the controls on the color page OR you can render-in-place the source clip which will then put its first frame in alignment with the first frame of the matte clip. I realize this all sounds like a crazy amount of effort but it really isn't. I find using the rendered matte in Fusion to be the simplest way of using an external matte but Chadwick shows us how flexible Resolve is with 3 different ways of accomplishing this task!

loses the mask for no apparent reason

It may not be apparent, but most often the mask is 'lost' because the source has changed. For example, changing Zoom or Position on the Edit page after creating the mask.