To clear up confusion, here is how it works
I've been using this program for years now. I don't use it for everything, but the way it works, it would allow you to use it for basically everything. There is a reason why you can use it for almost everything that comes as a frame from your GPU:
Lossless Scaling is like a screen capture program that intercepts the frames and manipulates them with whatever options you have chosen to manipulate them with. This is apparent when you try to record the extra frames from Lossless Scaling's frame generation (LSFG): The screen capture program can not record the extra frames from LSFG, because it grabs them for recording before LSFG can create its extra frames. That's why programs like Camtasia or the Windows Game Bar (that one might even crash the game with LSFG on) only record frames natively produced by the GPU. Since OBS can record your screen in several layers, it can record itself too, so with an extra layer specifically made for this purpose, you can actually record the output LSFG produces (there are how-to videos on youtube to make that happen if you are interested to do that for whatever reason). But in any way: It's not magic, and this way to grab the frames from the GPU is among the reasons why it increases frametimes (aka input lag).
The lag problem
This is the price to pay for increased frame rates. While the upscaler of Lossless Scaling (LS) does not actually make it much worse frame time wise, the frame generation part of LS does. It's not just that the frames the GPU produces need to be hijacked for manipulation by LSFG, the GPU also has to wait until the LSFG process is finished and has inserted the extra frames after the last native frame produced by the GPU, before it produces the next. That's why LSFG put to x2 frame generation is faster than x3, which in turn is faster than x4. While four times the framerate sounds great, everything that exceeds 20 ms of frame time will begin to feel sluggish. And depending on how intensive the calculations are your graphics card has to make in order to produce a frame, it gets increasingly likely to exceed those 20 ms frame time the more frames you set LSFG to generate.
Also, I never use x4, because it really is overkill for basically every game and the more frames that are being generated the higher the distortion problem rises. I have seen players complain that LSFG looks aweful when they lock their framerate to 15 or 20. You can test that yourself on how that looks (it looks badly, never lock your fps below 30 for LSFG to use). LSFG needs some data to work with, and the more frames you give it to process the less distortions and input lag it produces. To mend the input lag, use LSFG with as high a native framerate your GPU can muster. Distortions will be low to not noticable, and the lag time (because of the increased base framerate from your GPU) will go down too that way. Example: 60 fps as base are 16.6 ms frame time (input lag barely noticable for the average player). LSFG 2x will increase this to 20ms frametime while effectively doubling your framerate (so that's a good trade millisecond wise). If you increase that to 70 fps base (14.3ms) this will result in ca 18.6ms for double the framerate with using LSFG (etc.).
Yes, it's not an optimal solution and subpar to direct implementation like DLSS frame generation or FSR 3.1 (all of which have access to game data like motion vectors or ZDepth channel - data LSFG does not have access to), but if that is not accessible to the game you want to increase framerates with, LSFG is the only viable option to use instead. But the program is useless for competitive or e-sports players. Since the LSFG frames are basically a fake, they don't improve frametimes, on the contrary: They make them worse. So you will have a disadvantage using this program during a competitive gaming session (just don't use it for that).
Also also: If you have access to DLSS upscaling in your game: Use it. LSFG uses every input frame to generate new frames, regardless of source. So it will happily put frame generation on top of DLSS upscaled frames from the game. DLSS is the top upscaler, unrivaled by almost every other option. LS has it's own "propriety" upscaler (LS1), this upscaling algo is very good - it even approximates the quality of DLSS. So if DLSS is not accessible to you, LS1 is the next best option to use instead (it's better than FSR or Xess). It jitters a little more and is a little blurry in comparison to DLSS, but it's much better than the rest of the upscaling bunch.
Distortions
Since LSFG has no access to internal game data, distortions will happen. They implemented some "detection" method to preserve HUD elements, and comparitively to earlier versions this works really well, but it will jitter. In part even a lot, depending on the game HUD. In Cyberpunk it's there but barely noticable. In games like "The Planet Crafter", which has high contrast HUD elements as well as textures with high frequency details, this jitters a lot. It regenerates with every native frame coming from the GPU, but let's face it: LSFG is not magic, it can not descirn HUD elements from texture elements very effectively. If the jittering is a deal breaker for you (and I can understand why, for some games it's just too distracting) then LSFG is not an option for you to use. General note: The more native frames the GPU produces the less distortions will happen.
Can it be used for every GPU?
Depends. If the GPU was produced after 2011, then: Probably yes. Since most GPUs in use today were produced after 2016, I think you are likely to be good to go.
Is it for games only?
No. LSFG uses every source to generate frames with (like mentioned above: It works like a screen capture program; it does not discriminate where it gets its data from) - even videos. If you have the program: Fire it up, start a youtube video of your choice and then activate LSFG with whatever frame generation mode you like. You will see a 24 fps video stream will be at 48 fps with x2, 72 fps at x3 (etc.). I personally don't like to increase the framerate of hollywood movies, or certain Netflix shows, because there is a special quality to 24 fps with those. Increasing the framerate of those videos makes them look like being shot with a cheap videocam (because of the increased fluidity of the videostream). But I guess that's a matter of taste. In any way: LSFG can be used with every source that feeds it frames to work with.
Does it work with Steam games only?
No. It works with every source of frames coming from the GPU. You only need Steam to get LS, because that's the only place the creators of this program sell it through.
Are the "NVidia" and "FSR" options in LS like DLSS and FSR upscaling?
No. As I understand it, they are "approximations", but they are qualitatively below the native propriety upscaler of Lossless Scaling named "LS1". I dunno why they implemented those, but they are quite frankly pretty bad in comparison. If I upscale with LS I only do so with LS1, it just musters the best result. There are other options (also for the renderer used) for certain other cases, like 2D games, but I don't use them, so I have to refer you to the program's documentation on how (and when) to use them.
Closing notes
To get the best results from the frame generation of LS only: Let your GPU produce as many frames as it can. They changed the program with the last update in a way that you do not have to lock your framerate anymore to get an optimal result (it is still recomended to stabilize frametimes though), but with Herz manipulations like G-Sync or FreeSync an uncapped framerate works fine. Even if you don't have G-Sync and such, you still can simplify the process for using LSFG by just unlocking your framerate and set a few options for frame generation. The recommended minimum Herz your monitor should have to display the extra frames is 100 Hz. It seems a little low to me, but the more the better. It's not usable for any gaming situation there is, but if any of the in-built frame generation options are not available to you, LSFG is the best thing to use for that same purpose.
Alright, I hope that helped answering the most pressing questions regarding this neat little program.
Tristan
