Sony premier studio ND using a proprietary, purpose built engine for a flagship title with a far bigger budget and all the resources and technical expertise as needed provided by the platform holder vs. Remedy using UE...
Slight correction: Remedy used Northlight for Alan Wake 2. It's their own, purpose built engine.
While I certainly don't disagree with the video, I think it's worth considering
why this kind of change occurred. Looking at earlier 3D titles, like Quake 2, they actually have bounce lighting - albeit, quite primitive in their calculations. This type of calculation wasn't feasible in real-time, so they used pre-computed lightmaps and were able to use fledgling GPUs to merge the pre-baked lightmap textures with realtime lighting passes - things like muzzle flashes and rocket trails - to create the illusion of interactive light. As computers got faster and GPUs stepped up, we made it to fully real-time lighting - however, the sheer
performance cost of this type of approach was often prohibitive for the final result relative to what was still achievable with pre-baked lighting. As scenes, materials, and lighting got more and more complicated, the baking requirements in terms of time and hardware grew and grew. Moving to real-time lighting mean taking a hit in the visual arms race that developers engage in, so pre-baked techniques persisted. Oftentimes, games would need to contain half a dozen pre-calculated lighting passes for their entire environments and materials to create various lighting transitions, all to fake large-scale high fidelity real-time lighting. This baking process could take literally days to actually render, leaving artists and designers with no real way to see their work in-progress. This slowed down production as developers pushed these techniques to their limits. Slower production costs more money.
By the time we got to CryEngine, where the editor rendered everything in real-time, production was able to move much faster. More iterations, higher quality, less time. This pushed a lot of engines to move in this direction, because it allows developers and artists to see their output immediately, as opposed to needing to wait potentially days to see that you need to do another pass to get everything looking as you wanted it to in all scenarios. Faster production costs less money.
What we see with titles like TLOU2 isn't sophisticated real-time lighting - it's just what can be done with the hybrid techniques in 2020 using a $250m budget and literally thousands of artists to pull off the illusion. Even Naughty Dog's smooth animations are actually pre-baked, using a technique called motion matching to merge together thousands of pre-canned animations. The sheer amount of manpower necessary to produce stuff at that kind of scale is prohibitive for everyone except those with the biggest budgets in the history of the medium. For their artists, they usually create local-rendered versions of their work, just rendered at significantly lower fidelity before they can merge it into their nightly builds. This provides them at least some ability to check their work, but the process can still take several days to final something that could take only a couple of hours if everything were real-time. The trade off is their games look that much better than everyone else's because the machines are calculating that much less every frame. For a first party looking to deliver system-selling visuals, that's money well spent for Sony.
In order for real-time to keep up with pre-baked techniques, something has to give - and in this case, it's usually image quality and performance. Realtime will look just as good as pre-baked when rendered at native 4k on a bleeding edge dual-GPU PC, but when upscaling from 600p using FSR2 on an Xbox Series S, it runs and looks like vomit soup. The trade off is made so that developers can work faster and easier, and change jobs on demand due to standardised technology like Unreal Engine, all working to help games cost less money to make. For example, the biggest selling point of Unreal Engine 5's Nanite wasn't the lack of pop-in for gamers, it was that developers no longer needed to massage LOD systems or have artists hand-craft LODs for games using the system. Running your game at 30FPS instead of 60FPS, but needing one year less of development costs, is an easy trade off to make in this industry.