Just How Can L.A. Noire be Improved for Current Generation Consoles?

Discussion in 'Design and Criticism' started by GaemzDood, Oct 9, 2017.

  1. GaemzDood

    GaemzDood Well-Known Member

    L.A. Noire is easily one of my favorite games of the last generation, but given the numerous technical problems it suffers from (poor framerates on last gen consoles, other issues on PC) in addition to just how amazing Rockstar's RAGE engine can look, I would like to see them push even further than Grand Theft Auto V.

    1. Subsurface Scattering, Realistic Ambient Occlusion, Bokeh Depth of Field, etc.
    L.A. Noire's animations were impressive, but the shading wasn't. Light and skin interaction was lacking, especially compared to other games at the time like Heavy Rain.

    This is what I mean.
    (Ignoring the atrocious ambient occlusion, lighting and materials have always been a high point of Quantic Dream's games from Heavy Rain onwards.)

    Eugh. Also note the low resolution materials that basically lack any light interaction.

    Character rendering has massively advanced since the release of both games, and with techniques like Separable Subsurface Scattering which uses high-quality separable approximation that is provably optimal for additively separable irradiance signals, this should be used for the remaster.

    Improving character rendering is really fundamental for this remaster, since that is the catch to LA Noire, is it not?

    Take notes from GTA. The character rendering and post processing differences between the PS3 and PS4 versions are quite literally generational.

    Also, if you've played LA Noire, you probably noticed the cartoonish SSAO shading or the depth of field halos. Weigh the samples in a depth buffer for the remaster, and for both the depth of field and SSAO, preferably switch to a non-linear blend. And in the case of SSAO, point that shit upwards. Most light comes from the sky, and indirect shadowing should represent that; it is far more realistic than adding SSAO around objects. Also, look into the Monte Carlo Ground Truth AO, as used in recent COD titles.

    This is what temporally stable depth of field looks like. Notice how un-gamey it looks.

    Also, temporal supersampling does wonders.

    2. Lighting Improvements
    As I stated in my Fallout 4 thread, adopting a deferred rendering pipeline is pointless if you are just going to use point lights in places were objects should cast shadows. Rockstar understands this, as if you've played Grand Theft Auto V on PC with Shadows cranked to Very High, headlights cast shadows on everything. However, that should apply to more things, such as muzzle flashes, particles, god rays, and what have you.

    Also, another pet peeve with bad deferred renderers is when there are tons of direct point lights everywhere without any indirect lighting. Irradiance probes, as seen in Assassin's Creed IV: Black Flag, works well with dynamic light sources and dynamic TOD/weather systems.

    Also, import the fog system from Grand Theft Auto V on PS4/Xbox One/PC. Fog in that game was actually transparent, worked with volumetrics, and interacted realistically with light. Problem is that it caused edge dithering due to the low resolution, which can be solved through temporal reprojection/resampling. Hell, now that I think about it, GTA V has some of the best non-physically based lighting I've seen in a game.

    3. Anti-Aliasing Improvements
    For years, gamers clamored to MSAA, but after DOOM's 8x TSSAA or Call of Duty's Filmic SMAA, I'm content in saying that temporal supersampling has won. Grand Theft Auto V attempted to stick with MSAA and offered up MSAA with temporal filtering [TXAA] as the next best option to tackle temporal aliasing, but due to the way deferred and HDR pipelines work with MSAA, it often caused shader aliasing and subpixel break-up, even with TXAA enabled.

    LA Noire's remaster should go the route of temporal supersampling, especially for the portable 720p Switch screen.

    Side Note: If you're trying to incorporate MSAA into a deferred pipeline, perform a custom resolve (preferably with a wide Gaussian filter) post g-buffer and create a subsample mask. For pixel frequency passes, reserve bits of stencil read mask for per-pixel regions and bind pre-resolved (non-multisampled) targets SRVs. For sample frequency passes, set the stencil read mask to reserved bit for per-sample regions (0x80), bind multisampled targets SRVs, and index current sub-sample via SV_SAMPLEINDEX. And please use your own coverage mask (like Crysis 3) as opposed to traditional alpha testing (like Grand Theft Auto V), as the regular route only covers triangle edges. Also, if you want it working with HDR, read up on Ready at Dawn's Order: 1886 presentation for how they fixed it.

    4. Textures, Materials, etc.
    Grand Theft Auto V's parallax occlusion mapping was extremely well done. It was subtle, stable, and had so much detail that even the displacement mapped rocks would cast shadows.
    If you look at the original release of LA Noire or Grand Theft Auto V's 7th generation version, materials were low resolution, barely defined, and lacked proper material interaction. I'm not saying rewrite all of the textures to support PBR, because that can end badly in remasters (Darksiders II; although to be fair, that game wasn't targeting realism and their BRDF didn't reflect that), especially when they're coming out this soon, but add depth to them via realistic specular and light interaction, raise the resolution to support 2k since the game is 1080p (4k textures for Xbox One X and PC), add crests and details that were not there before, and make light realistically bounce off textures.

Share This Page