Hacker News Reader

Comment by rcthompson | Hacker News Reader

Everyone commenting here that they don't care about photorealism in their games is missing the point. The example games (and movie) given at the start of the article lack an appearance of photorealism not because they were going for something else, but because they either didn't pay enough attention to the HDR rendering, or else they didn't have the expertise to do it right. The article also gives examples of games that intentionally went with a non-photorealistic look to achieve a specific effect. So the point is not "these games look bad because they aren't photorealistic", it's "these games look bad because their HDR rendering is bad".

I don't know enough myself about HDR rendering to have an opinion on whether the author is right or wrong about any of this, but people arguing that the author is wrong because games shouldn't look photorealistic are attacking a straw man.

View on HN

The author doesn't appear to have bothered to investigate or honestly ask WHY games are the way they are, though. They are just blindly attacking an entire industry on basically the claim that the entire industry is stupid.

But the industry isn't stupid, and some of the "over-contrasty garbage" is intentional to help people rapid identify the things that they need to identify. Being able to quickly spot the things that you need to play the game are more important than sensible tone maps and contrast curves.

This is not a medium you observe like film & paintings are, it's a medium you interact with, and rapidly at that. Approaching the visuals with the same perspective as something you only observe is a fundamentally flawed approach.

This article is packed with baseless assertions, unfounded claims, and offhanded but not investigated technical limitations. Hell, right after saying "partly due to technical limitations" the author proceeds to throw out a "nobody in the game industry has an understanding of film chemistry" snark. The author doesn't even bother to elaborate on the technical limitations that they remarked on, opting instead to just hurl insults for no reason.

I was hoping that the author would go into how to do tone mapping correctly.

> ask WHY games are the way they are, though.

To add to your point, the main reason for "bad HDR" in FPS games is for simulating eye adaptation. Screens don't have 1,000,000:1 constrast ratio and can't shine as bright as the sun (we wouldn't want them to). This means that games are forced to approximate this effect - inevitably resulting in shortcomings.

Furthermore, game development is a race against time. You have milliseconds to render the scene and apply tone mapping. It may very well be mathematically possible to create automated tone mapping that would make a cinematographer proud, but can you make an algorithm that completes in less than 1 millisecond? I doubt it.

To my point, there are mods on PC that replace the tone mapper (I think ReShade does this) and they absolutely destroy your framerate - however the results are absolutely gorgeous if you know what you're doing.

> To add to your point, the main reason for "bad HDR" in FPS games is for simulating eye adaptation...

How is that different from what photographs do?

> Furthermore, game development is a race against time. You have milliseconds to render the scene and apply tone mapping.

Maybe I know just enough to be ignorant of the topic, but LUTs are just lookup tables. People working in film apply LUTs to the footage they're watching in realtime on normal PC hardware. Fragment shaders have been common in games for years. Last year I was working on a VR project (targeting 11ms per frame) and I think the post-processing took between 0.25 and 1.5ms--any more an it would have been dropped). For a desktop game having 16 or 30ms to render many fewer pixels is an eternity, comparatively. The only difference between a LUT and ICC profile (which most OSes support out of the box) is that for ICC profiles are mean to be calibrated against either an input or output--they're used everywhere in realtime.

I've never actually used ReShade, so I had to look it up. It does way more than color correcting. Depth of field, ambient occlusion, antialiasing are completely separate from tone mapping and notoriously computationally expensive (not just in video games, but in film as well). I also think since third-parties are writing them, they're not always tuned to be performant.

> How is that different from what photographs do?

A typically developed HDR photo is a single sample following the time at which the eye has adapted. There are [hopefully] no over- or under-exposed areas.

The dynamic range in HDR games is often used to simulate the entire process of eye adaptation. It's a function over time that intentionally creates over- or under-exposure some of the time.

I still don't see why animating the exposure explains why there's "bad HDR" in games (high contrast and over saturation)? I haven't implemented it or read any papers on the subject, but the games I've played don't look that much different than auto-exposure on camcorders in the 90's (or your phone today). For actual cameras it's more difficult because you need to guess and adjust the aperture or shutter instead of having a HDR image in a buffer and deciding how to generate a LDR image to present. Calculating the exposure (especially if you're animating it and only willing to go so far in one direction or the other), should be fairly quick.

I recognize the author of this post just from the logo on his site. He worked in the industry and has spent years sharing his knowledge on gamedev.net. This is not an attack on the industry. He is speaking to his peers.

> [...] some of the "over-contrasty garbage" is intentional to help people rapid identify the things that they need to identify. Being able to quickly spot the things that you need to play the game are more important than sensible tone maps and contrast curves.

Is this speculation or were you involved in these discussions? Or, maybe a GDC talk? For all I know, you could the lead graphics programmer for Naughty Dog, in which case I'd totally take your word for it.

And, actually, I really would like to hear more specific criticisms. As much as I respect Promit, he absolutely could be wrong and a detailed breakdown from someone who knows better would be informative.

Whatever reason modern games have for looking like shit, it's not good enough. The end effect is that they look like shit.

Maybe a little snark is good. These big studios have been pumping out bigger and louder games for a decade, and each one is more soulless than the last. There's a reason Minecraft took off the way it did. It's because it was a return to the visually appealing age when pixel colors were in some sense manually selected, resulting in a pleasant palette. I can say without a hint of irony that I prefer the look and feel of the original Doom to any modern blockbuster game.

Are you here to tell me that Minecraft and Doom are paragons of bad gameplay?

Maybe I wasn't clear. I said, basically, gameplay trumps visuals. You are in agreement that gameplay trumps visuals. If gameplay and visuals are at odds, gameplay should win. Minecraft's contrast is boosted and its colors are not accurate, and this all adds to its charm. This is in dispute of the article, which is taking the stance that tone mapping and colors should be tuned to look like a painting or film, rather than aspects that can influence gameplay.

My response here is mostly that many of the visual choices made in the games I'm criticizing were made completely independently of the game design team. This is far from ideal but it's a consequence of how the production pipeline is set up in these blockbuster titles. I do vehemently disagree that the aesthetic traits I'm looking for are at odds to gameplay and design decisions, though.

I could go on for a while about how Zelda BOTW manages to integrate the design and aesthetic choices, but I'm going to get mugged if I try to use that game as an example again. Maybe I'll use it as part of a different write-up on visual design.

Zelda BOTW has climb-anywhere gameplay, though, so it doesn't need landscape contrast to support gameplay purposes. Contrast that to something like Horizon Zero Dawn where you need the harsh ledge contrast on rock faces to distinguish between walls you can climb on and those you cannot.

Similarly combat in BOTW happens up close, you don't actually need to distinguish things at range. This is not generally true, especially for the other titles you're comparing BOTW against. BOTW also has problematic cases such as this: https://static.gamespot.com/uploads/scale_super/1552/1552458... That rock face is horrible. Totally washed out in a highly unrealistic way, to say nothing of the ridiculous saturation of the torch light.

The game industry definitely follows inward looking trends. For a long time every game was grey or brown and lacked contrast. More recently they all seem to use the same look/LUTs.

I worked in animated feature films and had similar critiques about the lighting and look. I felt like they lit everything like a well exposed photograph and were afraid of letting things fall off into dark or light (part of that could have been poor simulation of the toe or head of the film curve).

Then, when Pixar did Wall-e (2008) they hired a well known cinematographer, Roger Deakins, as a consultant. DreamWorks also hired him for How to Train Your Dragon (2010). I think that was a big change and both of their studios approached how CG movies look.

Or, you know, the developers were trying to balance far more factors than what has to happen with a movie. Multiple platforms, not only high FPS but maintaining that FPS with disparate scene complexity, old hardware vs new, support a variety of camera distances and fov, highlighting game elements so that objects don't blend together, handling in-game cutscenes as well as actual play, having an asset pipeline that doesn't destroy your budget, etc. Not to mention players that are random monkeys that like to try to do bizarre things within your game.

I guarantee game developers and game artists know a hell of a lot about HDR and color and lighting, and all kinds of things related to it, probably more than most people involved with movies. They just come at it from a different viewpoint - that of real-time rendering.

I can confirm everything except gamer audience concerns. I don’t game. I have worked in film production camera department, post-production and now CGI. Color grading and compositing in particular are not specialties of mine, but I know the systems well, and I can confirm the technical critique in the text. This author does an impressive job of cutting out the bull and getting to the bare issues.

As a gamer, I can also confirm that I tend to have a tough time with dark areas in a lot of mainstream games. If I can get the game to realize that I'm trying to look at them (usually by getting the bright sky out of my viewport so it does the "HDR adjust" fake thing) then the darks come out of the contrast floor and I can suddenly make out the detail.

The problem with the game doing fake-HDR is that while it looks great in trailers (or... at the very least, looks like the engine is doing something fancy with lighting, "great"-ness notwithstanding) it doesn't match what my real eyes would do in a situation. Unlike film, a video game doesn't have control over where I'm going to point the camera. I do. And sometimes, I need to see over that ridge, but the detail in the shadows is important too, and the game camera can't see what I'm actually looking at (with my eyes) and adjust accordingly.

Breath of the Wild is a great counter example, because it doesn't really try to do the HDR adjust thing at all. It doesn't really need to; sure its daylight scenes are a bit washed out, but real daylight is a bit washed out too. My eyes adjust, just like they would outside, and I can actually see what I'm doing. That's important! With games, tempting though it is to control the camera effects just like you would when shooting film, you can't control the player's actions. It's usually better for gameplay to mellow out the effects, so the player can focus on their actions without being distracted by the 'pretty' effects.

> The problem with the game doing fake-HDR is that while it looks great in trailers (or... at the very least, looks like the engine is doing something fancy with lighting, "great"-ness notwithstanding) it doesn't match what my real eyes would do in a situation.

An especially egregious offender in this category is Battlefield 1. When looking out of the driver's hatch of any vehicle, the game will randomly and chaotically fade it's "HDR" simulation between the exterior being discernible and a white rectangle (the hatch) set on a solid black background (the vehicle interior). The fade depends on miniscule adjustments in the PoV of the player, which is fixed relative to the outside world, not the vehicle, so you have to constantly track the view port.

Battlefield 1 generally looks best in scenes featuring low (natural) contrast.

> Yeah, and we screwed that up too. We don’t have the technical capability to run real film industry LUTs in the correct color space

(The reason being that even fairly simple color correction pipelines in a real color grading tool keep a mid-range GPU pretty busy at 1920x1080 and higher resolutions).

> I tend to have a tough time with dark areas in a lot of mainstream games. If I can get the game to realize that I'm trying to look at them (usually by getting the bright sky out of my viewport so it does the "HDR adjust" fake thing) then the darks come out of the contrast floor and I can suddenly make out the detail.

Of course, this is also how eyeballs work...

This is a valid defense! The real issue though, primarily, is that video game cameras don't emulate this well enough. My eyes don't respond nearly as fast to light changes as a video game camera does, and they're not as sensitive to small variations in the overall light level. They're also quite a bit more sensitive, especially to darkness and shadow, even in bright daylight. I find that most video game lighting exaggerates the shadows way too much.

So, adjusting the curves, especially at a slow, steady rate? That might be a good idea, and should actually aid visibility if it's done right. Say, walking all the way into a dark cave should lighten things up a bit, since your real eyes (and most cameras) would do exactly that. On paper, this part of the HDR effect works, and if it's done well it enhances the gameplay rather than distracting from it.

Imitating a faster, more jarring camera transition when it fits the mood, especially during cutscenes? That's fine, and the article gives some great examples of this being done well. Scaling up the contrast so that any HDR changes cause you to either lose all the detail in all of the darks, or all the detail in all of the lights? That's the problem. The effect, if it's there at all, needs to be a lot more subtle. This is partly due to limitations with the final RGB space (only 255 color values to work with and weird curves to boot) but mostly due to the fake-HDR effect just being a poor imitation of nature.

I am a colorist and I also want to add that the technical details check out - minus the Red One being better than film. It was most decidedly worse in every way - but was almost as good which was the important thing. I recently was going back and look at One and MX footage and couldn't believe how terrible it has held up while the Alexa sensor is still great today as are the 16 and 35 scans I work with.

That aside, I found the discussion about the misuse of ACES particularly interesting. It's like the completely misunderstood the tool and how it's intended to be used (which isn't uncommon with the tech unfortunately).

Yes, the bit about what the RED One was trying to do was a bit cringy but I thought it skirted around claiming it was better than film. The comparisons between the Alexa and RED One were very good, however. Coincidentally, I was doing work for Oakley as the RED One was being tested. After two test shoots with it, the director declared it B-cam and let a PA operate it, only because we couldn’t get the color right. This was action sports and the frame rate would have been useful. Even now, I am not sure much has changed. I don’t go on sets or handle much live footage these days but my observations have it that much of the production industry who could use it still lacks understanding of ACES. I have not worked with Alexa on any projects where I interacted with both camera and footage but my understanding is ARRI remains the king of blacks.

Reply on news.ycombinator.com