The maze is one space that re-skins itself as the player moves through it. Each instance is a new pass through the same geometry, but the further in you go, the more the walls absorb props, lighting and weather from places you've already been. A hospital door appears inside a corridor. Streetlight comes through a window that wasn't there before.

Previously used portals become destroyed as the maze progresses — the exits you relied on last time are gone, forcing you deeper. Elements from the hospital and childhood memories invade the geometry, and large surreal objects can appear in the sky where the sky used to be.

We want the maze to feel increasingly unstable without announcing that it is unstable. The player should recognize that something is wrong before they can name what. We're keeping the rules of the collapse deliberately opaque — if it can be gamed, it stops being a memory and becomes a level.

In the hospital, the player never stands in the corridor. They sit in a monitoring room they cannot leave and switch between security cameras to navigate the character through the building. Each feed has its own angle, its own delay, its own damage. Some freeze. Some loop. Some show a version of the hallway that doesn't match the others.

The central design problem is orientation. When the camera cuts, the player loses their sense of where they are and which way the character is facing. We're building the switching system so that each cut teaches the player to read the environment from a new angle rather than punishing them for being disoriented.

Several puzzles require combining information visible only from specific cameras — a door visible on one feed, a code only legible through another, a movement pattern you can only predict from a high-angle shot. The player becomes an interpreter of incomplete footage.

The vulnerability is the point. You have access to a surveillance system, which should feel powerful, but the system is broken, partial and not built for you. You see more than a person should, and less than you need.

One puzzle asks the player to discover a numerical code by opening and closing bathroom stall doors until the correct combination reveals the digits through a mirror visible on a single camera angle. The code is always there, but only one arrangement of doors makes it readable.

Another requires restoring power to a sliding door by navigating to an electrical panel, interacting with circuit breakers and fuses, then switching back to a camera that overlooks the door to confirm it has opened. The puzzle is split across physical space and the camera network.

We're deliberately avoiding puzzle machines — contrived devices that exist only to be solved. Every puzzle in the hospital is built from things that belong in a hospital: doors, panels, fuses, mirrors, stall locks. The logic is the game; the furniture is real.

The goal was to make the puzzle feel like something that could realistically exist in an abandoned hospital rather than an artificial puzzle machine dropped into a corridor. The electrical room has a history: labels worn half away, breakers that haven't been touched in years, fuses that have to be found and carried from elsewhere in the building.

We built the panel from references of real hospital infrastructure — the kind of equipment that lives in basements and maintenance corridors that nobody photographs. Props, panels, fuses and wiring are modeled close to their real counterparts because the believability of the puzzle depends on the believability of the room.

Scale is one of the tools we use to shift a memory from familiar to wrong. A distant object that is simply too large — a shape on the horizon that shouldn't be there at that size — does something to the body before the mind catches up. It's a quieter kind of horror than a monster.

We're still testing how far to push it. The effect depends on contrast: the object only feels wrong because the rest of the environment feels grounded. If everything is surreal, nothing is. The impossible things have to intrude on the real things.

The factory is roughly 9000 square meters split across multiple levels: large open manufacturing floors, storage areas, offices, changing rooms, a cafeteria and industrial machinery. The layout is a single continuous loop so the player can lose a pursuer, change elevation, and come back around without hitting a wall.

We deliberately avoid too many dead-end rooms. During pursuit the player needs opportunities to escape or change direction at almost every turn. A dead end in a stealth game is a death sentence, and we want the fear to come from being chased, not from being trapped.

Sound carries through the metal structure, which means the player's own movement is what betrays them. The factory reads as navigable, logical, almost safe — and then the lights go out on a schedule nobody gave you.

The pursuer is dressed in black — a black hat, a high-collared dark coat, the face hidden in darkness. We're keeping the character's identity and motivations out of these notes. What we can talk about is behavior.

The AI is built around a perception model: visual perception with line of sight and viewing angle, and hearing that responds to the player's movement and the surfaces they move across. The pursuer can pursue, investigate, search, lose the player and reacquire them. It doesn't know where you are. It works it out.

The goal is a pursuer that feels like it's reasoning, not running a script. When it loses you, it searches the last known area. When it hears something, it investigates. When it sees you, it commits. The pressure comes from not knowing which state it's in.

The sequence is built around movement, atmosphere and pacing rather than combat or dialogue. A child on a bicycle, a vehicle behind them, a street at night. The gameplay is about getting away — reading the road, choosing turns, feeling the distance close.

The sequence transitions between perspectives. The transition into the factory is earned by the chase, not announced by a cutscene.

Environmental storytelling carries the narrative weight here — the street, the weather, the things passed along the way.

The apartment building is a nighttime investigation. The player moves through corridors, past locked doors, past residents who saw or didn't see something, past environmental details that don't quite agree with each other. A police radio communicates context the player has to interpret.

Blood traces, clues and environmental details are placed to contradict each other rather than confirm a single narrative. The investigation gameplay is about reconstruction — building a theory from fragments that resist becoming a single story.

We're avoiding spoilers here. What we can say is that the building is designed so that the more carefully you look, the less certain you become.

Most games pick a camera and keep it. Forgotten Self doesn't. The perspective shifts depending on which memory the player is inside, and the shift is not cosmetic — it changes how the player reads the space, how they move, and what they're allowed to know.

First person is for close observation. Third person is for pursuit and movement. Second-person surveillance is for memory that has to be watched, not lived. Investigation is for reconstruction. Cinematic sequences are for the moments the memory refuses to be interactive.

The perspective is part of the narrative language. When the camera changes, the memory is telling you something about itself.

The visual identity is a contract with the player: this place is real. The environments are grounded — abandoned hospitals, industrial spaces, apartment corridors, nighttime streets — built from reference and lit like the real thing.

Then the surreal intrusions arrive. Analog video damage, surveillance monitor color, objects that don't belong, scale that doesn't add up. The intrusion is always partial. The realism has to hold so the wrongness reads.

We're sharing concept art, storyboards, environment tests, props and lighting experiments as they develop. Not everything here is final — most of it isn't.

Sound in Forgotten Self is used to suggest threats that may not always be visible. A whistle heard at a distance, a change in the rain, a heartbeat that accelerates before anything appears — the soundscape is doing work the visuals aren't.

We're building positional audio so that the player can locate things by ear, and a heartbeat system that responds to the perceived danger of a space rather than a scripted trigger. The tension crescendos are designed to blur the line between reality and memory, so the transition between states is felt before it's seen.

Industrial ambience, distant thunder and rain are the baseline. Everything else intrudes on that baseline. The goal is a soundscape that feels like it knows something the player doesn't.