FEATURE REQUEST REALTIME FOR LOADERS

Please make something similar to optimized media from resolve, this is so handy that I even started doing simple masking and tracking in it cause ones the media is optimized nothing slows it down so I can track and set keyframes and color correct in real time in 4K evenly as in FHD! Ones the media is optimized it is very hard to get something not real time, you really have to have a lot of nodes of correction or layers of footage to drop the speed. I think this could be a long time ago in every compositor cause in it you only work one shot at a time, even if to render it to a specific format took 10 minutes, though this must be a gargantuan footage, I'd prefer to wait so once it's rendered there is realtime covered. It sould be the first thing to be - load up the footage wait some time and bam! it's realtime. constantly. until you over weight it with loads of nodes, but with a couple it must be realtime in 2016 don't you think?
I really find tracking masks with optimized media in resolve hugely timesaving. I don't know any other software that could track 4K that fast. But it is only because of Optimized media.

P.S. When you set to cache disk afterwards it still loads up into the memory first so cache disk isn't really halping.

So I plant the seed here.
I think this is the very missing feature in every compositor.

What're your thoughts on this topic?

-

Because I've tested them, well, not all but these ones in particular, Nuke 10 claims to have realtime performance but still, renders loaded 6K r3d file 1 frame per second, even if you set 1:8 resolution (which is opposite to Fusion(which is goodbut still not enough) that renders 15 frames per second in 1/8 proxy enabled mode)
Ae renders even faster, and it even has the proxy feature that allows you to enable/disable the proxy for the particular comp, and this is the closest but not realtime though. And none of them using this that resolve uses now, it is like transcoding in avid but smarter cause you might (and certainly you can) want to disable it and see the final image and then turn it back on again. I'm willing that compositors as well have this function just to fast scrab through the footage.
Well Fusion even has it when you alt drag the loader to the viewer and it renders a flipbook file that is temporal and it's in raw and meanwhile it plays fast, but what a pity you cannot just take it and start adding nodes on that superfast and convenient raw file - it's just for the preview purposes.

was is a simple composition for you?

i have a simple comp with now around 280 nodes (just some keying, 3d space background, some colorcorrection) do you expect this to be in realtime?

you can always render your media to fusion RAW files and use them in a loader. you can even build a script that will do that and replace the loader for you. but RAW are huge and dont support the aux channels, which makes them useless in many cases

i'm not against any speed improvements, but your request is very broad and playing back footage and doing some colorcorrection is not the normal case for a compositor

-

you can always render your media to fusion RAW files and use them in a loader. you can even build a script that will do that and replace the loader for you. but RAW are huge and dont support the aux channels, which makes them useless in many cases

Sure they do. They support everything that a Fusion image does, including metadata. Well... Not sure about customdata, and of course you lose things like concatenated transforms...

Well Fusion even has it when you alt drag the loader to the viewer and it renders a flipbook file that is temporal and it's in raw and meanwhile it plays fast, but what a pity you cannot just take it and start adding nodes

That's how ALL of Fusion works though. You cache one output, then add tool after it and you only have to render those tools once before they become cached. As soon as the frames are cached, you can do 6K at 60fps. Works for shaders too. The textures (and the whole scene) is cached in system RAM and in the GPU's RAM, so you can easily do 8K color correction (float16) using my shaders in realtime. Basically limited by your screen refresh rate. That's why the most important thing for a Fusion workstation is RAM. You want as much system RAM as you can get (256GB is affordable now) and as much GPU RAM as you can get (8GB+ is affordable now). Unless of course you want to just get a monitor with a faster refresh rate...

EDIT: So on a whim I did a little test... 8196*4320 image, float16. I can run 44 of my color correction materials on it and still get ~60fps. Next I'll see how fast Resolve runs with 8K and 44 nodes...

EDIT 2: Well, Resolve didn't fare too well. It can't play 8K @ 60 with 1 corrector. Doesn't matter if I use 1 GPU or 2 GPUs. So as far as realtime color correction performace goes, Fusion is at LEAST 44x faster than Resolve.

Guys I appreciate your knowledge here, but do I miss something in the settings controls?
From my point of view I mostly talking about preparation for compositing, like tracking and masking.

By simple compositing I expect some daily work on an edited material, adjusting some shot, stabilizing replacing takes. Just some basic things you could actually do in an NLE but just with lesser controls on compositing with blendings lights etc.

So I load up some Dragon footage and willing to track fast a mask to replace a face in a shot from one take to another, the shot was shot on a movi and the tracking is simple and solid almost linear though I even could animating a coupler of frames and be satisficed. But I still want the tracker to do it for me, so as I loaded footage in to fusion I start tracking and it gets really slow like around ten minutes for a shot of around 500 frames.
And in my case I have two takes to be tracked And it already takes more 25 minutes + 10 minutes I adjust lost tracking center as the object moves further.

All this against what I did in resolve- first it tracked the face in ten seconds with over the optimized media. then another 5 minutes I animated by hand the tracked face in a position of a face in the second shot. Bam! I'm finished.

Just because this was a resolve I couldn't copy the tracking from the second shot to get it fully automatic just because it's not a compositor after all.. And the job is done faster though with lesser precision.

I believe softwares such as Fusion concentrate on advanced stuff and forget about the simplicity of the basic things, It's like they say a statement "As I'm advanced in this aspect of work and I could do much tougher things, the workflow of simpler things would be that tough too cause I'm not supposed to be easy. So the Samaritans would be confused in using me and then I could sort some generous purposes users as well."
I've just thought this statement up but that's how I feel..

That's a different issue though. And frankly, Resolve makes things a lot harder through it's arcane UI and insistence on preset project formats. Doesn't help that you can't automate or customize anything either.

The reason Resolve is fast and easy is the same reason a drill press is so good at making holes. But if you want to lathe a spindle, it's pretty terrible.

Fusion is designed to be a perfect workshop. And in that it largely succeeds. You have resolution independence that just works. You can easily handle (as I just discovered) 16K float plates at your monitor's refresh rate. You get a robust SDK including Fuses. You get two (three depending on how you count) different scripting languages. You get commandline rendering, including an immediate mode. You get network rendering. You get (or used to, it's missing from the current beta) network clustering and the ability to customize your UI. You get nodes that support more than a dozen different datatypes in a single flow.

Is it missing some things? Sure. I'd like to see multi-GPU support. I'd like to see OpenCL caching working again. I'd like to see better local caching. I'd like to see some expanded SDK access. But overall it's closer to perfect than almost anything else.

The trick, however, is to think of it like a workshop and not a drill press. You want a 5 axis milling machine? Go ahead and buy one or make it yourself. Fusion is the workshop for the tools, not the tools themselves.

That's totally sweet Fusion been a workshop but it only counts when you have time to develop, experiment, and create some new useful tools which would make your primary job easier. But if you have only one evening and the shot mast be done by dawn. On one hand you have an after effects guy who knows how to mix some default tools to get the desired result in couple of hours, on the other hand you have the developer who has a workshop and he knows how to create the exact tool which is needed for this specific job, but guess who wins in the eyes of the client? While the developer's to debug his new tool the ae guy has already shown the final render.

Not really. You're comparing an experienced AE user to a novice Fusion user.

I've seen lots of cases that are the reverse. A simple task becomes difficult for an AE user but can be done easily in Fusion. A smart lead will know which tasks to assign to which compositor.

As you get better at Fusion, you'll start accumulating the tools you need to make your work more efficient.

I'm not saying that every task you want to do is even practical in Fusion, but the same can be said about other software. There's things that Fusion needs to have to attract and support developers, and there's things Fusion needs to provide the built in tools that do a good enough job for more situations.

If your request is "REALTIME FOR EVERYTHING AND MAKE IT EASY ENOUGH FOR A NOVICE TO GET PRODUCTION SHOTS DONE IN A FEW HOURS WITH NO PREPARATION" then that's a pretty broad request that is out of scope for basically anyone to attempt to address.

No, that's not my point. This realtime for everything stands just to have a great starting point, like in editing software, like in Resolve with optimized media enabled, tracking shouldn't be slow, scrubbing the footage shouldn't be less then realtime, turning of and turning of the nodes except the source shouldn't make it recache all over again, the source is the bottom, of all nodes applied to it why wouldn't it just cached one time, and never again and would play forward and backwards like in the flipbook in realtime.

BTW yes you can render out flipbook raw files and load them up instead of originals, but it doesn't do any good the y won't playback in realitme so there is no point in rendering them first and replacing the originals.

But Resolve's Optimized Media is an extra step which you aren't accounting for. It's not like you're commandline generating those proxies from Resolve while you transfer images off the mag, right? Furthermore, it's lossy. Going from an EXR sequence to a ProRes 4:2:2 1/4 res proxy isn't really useful for compositing. If you can deal with that (because you're just scrubbing through to find the right take or whatever) then fine, do that. I know a few people who routinely make proxied files for quick access to the right frames. If you load in 4:2:2 1/4 res compressed proxies into, you'll get a huge speed boost for I/O. Probably 10x.

Resolve doesn't use the same data that Fusion does. It cares about RGB 0-1. That's it. Basically what an editing software cares about. So tossing any extra data makes sense. But for Fusion, you have to think carefully about that. Because it's an extra step that can cause massive headaches when you discover you're missing important data, both image and metadata. And compositors expect that to be there. They care about every pixel in a way that Resolve and Avid users do not.

You can have both, though. If you use the ProxyFilename feature, you get some of the I/O boost, but the flexibility to choose when you use the full data or not. Your cache for that LD can be a mix of regular and proxy images and Fusion is perfectly cool with that, though I don't use that workflow myself, so I may be missing a corner case. Some users like more explicit control, though, so they use 2 LD's and switch between them as needed.

Of course, the other side of the question is how much of your comp is I/O bound anyway? Resolve doesn't stay realtime even with optimized media proxies if you add enough processing nodes to it. If your comp is only 15% I/O, then you can't get more than a 15% speedup even if you replaced all your LD's with BG's. If all it took to get realtime performance was faster SSD's or RAM drives, we'd all be using them. At least Windows and Linux users would.

Nuke as a compositor doesn't have any real-time features. Studio does; but that is not for compositing specifically.

Want to try an actual real-time compositor ? Check out TouchDesigner....

Well, I kind of checked it out, pretty amazing, but as far as I've digged in, It is very limited to export anything but mostly only whithin a mov container, am I correct?