Magic Mask effect disappears after re-opening Fusion

Hi everyone,

I seem to be having a problem with Magic Mask in the Studio version. Using 18.6.
Sometime I use Magic Mask to isolate something and it works just fine, but then when I go back to open the Fusion comp a few days later to make an edit, the Magic Mask doesn't take effect at all and I need to redo Magic Mask to get it to work again.

Thoughts?

The Magic Mask only goes to RAM when you generate it. If you want it to stick around, you should create a Saver and save it to disk. Then immediately create a Loader and add it back into your stream. Alternatively, you can right click and create a cache. Once you've done that, it should be there until either you change something on the node or upstream, or delete the cache for some reason.

Oh wow, I didn't know that!
Thank you!

The Magic Mask only goes to RAM when you generate it.

Not so, it's saved to a .PNG sequence in your disk cache folder. It's only cached in RAM if you view it.

Why is it designed like this? I have no idea. It's an incongruent design that, as we've seen, breaks people's comps. Disk caching for every other tool in Fusion is exposed to the user and controlled by them. Having a tool that not only secretly disk caches, but fails if you open the comp without the cache in the same (unknown) folder is unlike any other design in the entire application, and I'm concerned that this pattern may be repeated in new tools to the same end result.

It like they have a hard on for not using the most obvious workflows. Is it so hard to make a cache node? Not only that, but why use a png instead of the braw format that the rest of the caches use? It seems like somebody copied the Stable Diffusion depth map library over and didn't bother to add the file conversion from the output. Just pump it out as quick as possible, consistency and overhead be damned.

Anyway, just creating a cache still remains dangerous because you may lose it once closed and re-open.. I believe the Render in Place when in Resolve or export of the matte is more a solid way to fix the issue

How is writing the matte to disk twice less dangerous than just writing it once?

Haha, do you intensively use the Magic Mask ? Of course not... Because you'd have experienced a lost cache too, so what now ? Well, then you just stop trusting the MM cache and write, yes, a real solid matte which you can rely on.
Don't get me wrong; I'm writing about how the MM cache implementation specifically, not the overall cache principle in the software

So up to now, it's not about writing "twice", it's about choosing a solid WF which makes you sure you can find the same state when re-opening your project the day after. The most intelligent and solid ways to do this are Render in Place if you're in Resolve, OR exporting a bitmap matte - for example - in Fusion.

Thereby I persist : " just creating a cache still remains dangerous", it doesn't give any guarantee you'll find it the day after...
This is how...

How is writing the matte to disk twice less dangerous than just writing it once?

Because you have survived death twice as long, while the processes are in progress?

Haha, do you intensively use the Magic Mask ? Of course not... Because you'd have experienced a lost cache too, so what now ? Well, then you just stop trusting the MM cache and write, yes, a real solid matte which you can rely on.

Have you sent BMD a reproducible case for this bug?

Hey Chad
Actually, I’ve wrote in 80% of threads about magic mask, to help users with workarounds I’ve already tested.
BMD already knows the whole stuff, and even answered - maybe Peter, to be checked - that render in place should be the best solution up to now.
I am personally much more on fusion, thereby I prefer exporting a bitmap matte …
You can check on your own : making the cache and locking the node doesn’t help: you can’t make 100% sure you ll find it again if you close the project/comp.
In a train right now but you may easily find the topic where bmd confirms the best practices until a better update

I've seen caching done a few different ways over the last twenty years. Adobe uses a unified folder stash that is kept per project and has a set limit before being overwritten. Maya began with using localized caches per action exactly like Magic Mask. Those would get lost or overwritten or corrupted and you would have to resim your fluids. Then they moved to exporting the caches as external files from the main menu system. You'd then have to reimport from the file>import area to bring back in the alembics. Now they have finally given up all of that and do it the modern way. A cache node. You select the type of file and save it to disk. Then you check a box and it becomes a read node and you're back to work. Houdini has the same. Nuke also has a check box on their Write node that turns it into a Read.

You're both right. You shouldn't have to write it twice but I've had plenty of times where a node just blatantly ignores the cache that it has on it. I would say that going all the way out to the timeline and writing from there is a waste of time and a distraction to the process of compositing. Using a Saver Loader is so stable that I had forgotten that the cache button was even on the MM node. I gave up using that after moving computers to finish a shot and losing the cache because it was local to the computer but not the project.

Caching with a node is one of the most crucial steps to making Fusion a success going forward. It alleviates so many issues of memory errors, GPU overruns and frame glitches. It also keeps the proper workflow when using Resolve to be a complete end to end post production pipeline. You shouldn't break out Fusion to create a plate to be used in the same Fusion Comp. It's confusing and unreproducible for others that might join to comp later. You'd have to write notes next to your MM node explaining what you did. Don't get me wrong, it works. It's just a bit convoluted. Even if you're going to use the mask in color, all you need to do is add another MediaOut node connected to the mask. Then you're able to access it from a second In point on the Color page. A Saver Loader in the flow saved to a caching location in your project folder structure has been a standard for me going on five years. Never failed yet.