Review 05828 hours, instrumented
Cascade cost19 ms worst 1%
Score9.1 / 10
Instrumented review · 058

Lantern Quiet Is a Technical Showcase Built Around One Honest Idea

A stealth game where the lighting model is not presentation but the entire ruleset. We captured all twelve levels and timed every shadow cascade, and this is the rare case where an expensive system is expensive for a reason the player can feel.

28 hours played12 levels captured96 tracesScore 9.1

The idea

In most stealth games, light is a texture and detection is a cone. Lantern Quiet inverts this: what the guards can see is computed from the same lighting solution that is drawn to the screen. If a shadow falls across you on screen, you are in shadow as far as the rules are concerned, with no separate visibility volume approximating the effect.

This is not a new ambition, but it is the first time we have measured an implementation that holds up under adversarial play. We spent a considerable part of twenty-eight hours specifically trying to find a case where the drawn image and the simulated visibility disagreed. We found four, and all four are edge cases involving thin geometry at oblique angles.

What it costs

A great deal, and the game is candid about it. The shadow cascade update is the single most expensive operation in any frame we captured, and in the two largest levels it produces a worst-percentile frame time of 19 milliseconds against a 16.7 target.

LevelMean frame timeWorst 1%Dynamic light sources
Atrium13.1 ms19.0 ms34
Service corridors11.8 ms15.2 ms12
Rooftop approach12.4 ms18.1 ms27
Archive11.2 ms14.4 ms9
Loading bay13.6 ms18.8 ms31
Flooded sublevel12.9 ms17.3 ms22

The correlation with dynamic light count is almost linear, which is a good sign: it means the cost is where the developer says it is, and that reducing light count in options genuinely recovers performance rather than merely dimming the picture.

The lowest lighting preset removes eleven of the thirty-four dynamic sources in the Atrium and reduces the worst percentile to 15.4 milliseconds. It also, correctly, changes where the guards can see you. The game does not pretend that a graphics setting is cosmetic when it is not.

Where it breaks

  • Thin railings at oblique angles. Three of our four disagreements involved standing behind a handrail whose shadow appeared to cover the player and did not register.
  • Two glass material types. Our separate audio occlusion test found the same two glass materials misbehaving; in lighting they transmit slightly more than they should.
  • Fast light rotation. A swinging lamp at full speed updates visibility one frame behind the drawn shadow, which is detectable if you are looking for it and never mattered in ordinary play.
A rain-slick alley at night between brick buildings with a single caged wall lamp and steam from a grate
The rooftop approach. Twenty-seven dynamic sources, and every one of them a rule rather than an effect.

The design that earns it

None of the above would matter if the game did not build on it. It does, relentlessly. Guards carry lights, so a patrol is a moving change in the ruleset. You can extinguish, redirect and break light sources, and each of those actions permanently alters what is possible in the space. The best levels are essentially puzzles about editing the lighting solution into a shape you can walk through.

The Atrium is the strongest level in any stealth game we have played in three years, and its strength is inseparable from its cost: it is expensive because it is thirty-four separately meaningful light sources, and it is brilliant for exactly the same reason.

Who should hesitate

Anybody targeting a high refresh rate. This game will not hold 120 hertz on any configuration we have access to, and the lowest lighting preset that would help changes the rules of play. That is an honest trade rather than a flaw, but it is a trade, and a player who values frame rate above all should know it before buying.

Score

9.1. The highest score we have given this year, awarded to a game whose worst-percentile frame times are among the poorest in our recent tests. Those two facts do not contradict each other, and a review that could not hold both at once would not be worth publishing.