Four-agent workshop (Araminta, Ozzie, Miri, Gore) establishing visual
identity for The Settled Reach. 9 mood board images, 16 art direction
principles, 10 candidate decisions (D-042 through D-051), v0.1.1 sprite
plan. Post-workshop technical session resolved 3D render pipeline,
resolution chain (1024/256/64), and camera angle lock ("the angle").
Pipeline test samples included: Era 1 institutional wall + bar green
wall rendered through the full Nano Banana -> 3D model -> sprite chain.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
18 KiB
Round 2 Workshop Response: Araminta (Art Direction & Mood Board)
Agent: Araminta (Visual Designer / Q-003 Lead) Date: 2026-02-12 Workshop: Art Direction & Mood Board Status: Round 2 — Resolving Open Questions
Response to New Input: Tile-Based Base Building
This confirms rather than changes my Round 1 positions. Let me explain why.
My entire art direction thesis is: sprites are shape templates that the lighting system completes. A world composed of discrete 1x1 tile objects doesn't fight that — it IS that. Every wall segment, floor tile, chair, desk, crate, and door is a shape template. The lighting system makes them feel like a room.
What this changes practically:
-
The "environment" isn't a painted backdrop — it's a vocabulary of placeable objects. This means every object needs to be individually readable at 64x64. No detail hiding behind neighbors. Each tile sprite must communicate "I am a wall / door / chair / console" through silhouette alone, because the player (or simulation) places them on a grid and needs to parse the result.
-
Rimworld becomes a production reference, not just a readability reference. Rimworld's object sprites are the right scale, the right level of abstraction, and the right pipeline model. Individual items, top-down readable, tile-aligned. We add what Rimworld lacks: lighting, shadows, atmosphere.
-
Nano Banana's job description changes. Instead of "generate room illustrations," it becomes "generate individual object sprites at 64x64 with consistent style, bold silhouettes, and flat lighting (the engine adds atmosphere)." This is actually EASIER for AI consistency — single objects are simpler than scenes.
-
The three construction eras I proposed in Round 1 now apply to OBJECTS, not wall paint. Original-era furniture (standard-issue, institutional grey). First-renovation equipment (slightly different material palette). Recent additions (Lera's chosen bar furniture, newer consoles). The visual history is carried by the objects themselves.
-
Object density IS environmental storytelling. The bar feels different from the logistics hub not because of painted backgrounds but because of WHAT'S PLACED THERE. Dense, mismatched furniture = bar warmth. Sparse, uniform equipment = institutional efficiency. Exposed conduits with nothing on the floor = maintenance corridor truth. Rimworld does exactly this — a bedroom feels like a bedroom because it has a bed, a dresser, and a lamp, not because the walls are "bedroom walls."
This is good news for the pipeline. One object vocabulary, one style guide, consistent generation rules. The world is assembled, not illustrated.
OQ-01: Neural Insert Overlay — Geometric or Organic?
The proposed resolution path works. I'll accept it with specifics.
My Round 1 position was clean geometric. Gore's was organic thought-like. The synthesis is: precision rendering with organic PERCEPTION. Here's what that means concretely:
-
Data layer (underneath): Geometric. Grid lines are straight. Entity markers are clean circles/diamonds. Text is crisp. Connection lines between POIs are 1-pixel straight lines. The data is PRECISE because it comes from a computational lattice.
-
Render layer (on top): Soft bloom pass. A 2-3 pixel gaussian blur on the entire insert CanvasLayer at maybe 40% blend. The geometric precision is still there but it LOOKS slightly soft. Edges breathe. The grid lines have a gentle glow rather than razor sharpness.
-
The result: Technical precision that FEELS organic. The player perceives it as "natural awareness" (Gore's framing) but the underlying structure is clean digital (my framing). The lattice IS computational — but the character has worn it so long that the brain has naturalized it. The soft rendering communicates that naturalization.
Implementation: One CanvasLayer for insert data (geometric, precise). One shader on that layer: subtle bloom/glow pass. Adjustable bloom radius lets us tune the organic-ness. This is cheap in Godot 4.
What this means for the D-033 colors on the insert: The relationship colors (teal, green, amber, red) rendered through the bloom pass will have soft halos rather than hard edges. An amber entity marker won't have a crisp border — it'll have a warm amber glow bleeding slightly into the surrounding space. That actually reinforces Gore's "gut feeling made visible" argument. The amber feels like a vague sense of concern, not a data flag.
Gore was right about the emotional register. I was right about the production structure. Both survive.
OQ-02: Does the Visual Environment Shift When Conspiracy Activates?
I'm moving to the Gore/Miri position. Strict zero environmental shift.
Here's why I changed my mind. In Round 1, I proposed "5% cooler, deeper shadows" as subtle CanvasModulate shifts. But after reading Gore's "the absence of change IS the horror" argument and Miri's mechanical reasoning that environmental hints break the investigation mechanic — they're right. The game is ABOUT information asymmetry. If the station itself helps the player detect conspiracy, it undermines the core loop.
The proposed resolution path is correct: the shift is character-driven, not environmental.
What changes when conspiracy activates:
- Insert overlay density increases. More data points appear. More connection lines. More ambient processing flickers. The character's lattice is working harder because there's more to track. This is visible to the player as "my overlay got busier."
- Monologue frequency increases. More internal commentary. The character is THINKING MORE. This is a text/audio change, not a rendering change.
- Entity color changes. Green to amber. Amber to red. The D-033 system does the work. The information itself shifts, not the environment presenting the information.
What does NOT change:
- Light color temperature. The bar is still warm. The corridor is still dim.
- Shadow depth or direction. Same Light2D parameters.
- Ambient sound (that's Inigo's call, but from visual: no ambient visual change either).
- Tile colors, wall tones, floor patterns. The station is the same station.
The player walks into the bar after discovering the conspiracy and it's STILL warm and inviting. That's the horror. The warm light isn't ironic — it's indifferent. The station doesn't know what the player knows. It doesn't care. It continues being a perfectly comfortable place to live, which is what makes the player's knowledge so isolating.
One exception I want to preserve: The player character's OWN entity sprite could shift subtly. Not the world — the self. If the detective has been running hot (stress mechanic, if we build it), their sprite could show it in posture or a slight animation change. The character carries the conspiracy's weight visually. The station doesn't.
OQ-03: Entity Detail Level, Sprite Dimensions, AND Tile Object Vocabulary
The base building input actually resolves this question more cleanly than Round 1 could.
Here's the updated spec:
Structural tiles (walls, floors, doors)
- 64x64 per tile. This is confirmed from D-014 and the base building input.
- Structural tiles are the canvas. Muted, understated, differentiated by zone palette (Round 1 hex values) and construction era. They MUST NOT compete with entity sprites or object tiles for visual attention.
- Wall tiles need clear top-down silhouettes: solid = wall, gap = door, transparent = window/viewport. At this scale, walls are primarily information barriers, not decorative surfaces.
Object tiles (furniture, equipment, containers)
- 64x64 footprint per 1x1 object. Larger objects (tables, beds, consoles) can be 2x1 or 2x2 but composed of 1x1 sub-tiles for grid placement.
- Detail level: Rimworld's object sprites are the right target. Clear top-down silhouette, identifiable at a glance, flat-lit (engine adds lighting). A chair looks like a chair. A console looks like a console. A crate looks like a crate. No ambiguity.
- Object sprites must read WITHOUT color. Because in certain perception modes (thermal), objects lose their natural color. The silhouette alone must identify the object type.
- Objects carry the three construction eras through subtle material variation. Original-era desk (institutional grey, squared edges) vs. recent desk (slightly different tone, rounded edges). Not dramatic — a 10% palette shift within the same hue family.
Entity sprites (NPCs, player)
- 24x32 pixel footprint within a 64x64 tile. The entity is smaller than the tile it occupies. This creates a clear figure-ground relationship — entities float ON the tile grid, they don't fill it. Ozzie's Heat Signature ratio reference is right here: entities are clearly smaller than the space they inhabit, which makes the space feel like a PLACE, not a character portrait.
- Headroom matters. 24 wide x 32 tall gives a slight vertical elongation — human proportions, not chibi. The extra height means the entity reads as "person standing" not "token placed."
- Silhouette discipline (I-01). Each named NPC gets one identifying shape feature visible at this scale: body proportion (stocky/lean/broad), one accessory (vest/apron/badge/tool belt), posture (hunched/upright/slouched). Generic NPCs get 3-4 template silhouettes.
The entity-to-object-to-structure hierarchy
This is new and critical. With base building, we have THREE visual layers in the game world, not two:
- Floor/wall tiles (structure): Muted. The background. Communicates space layout and zone identity.
- Object tiles (furniture/equipment): Mid-detail. Placed ON structure. Communicates function and lived-in character.
- Entity sprites (people): Boldest silhouettes, D-033 color coding. Moving ON objects and structure. Communicates the social/information layer.
The visual hierarchy must be: entities > objects > structure. The player's eye goes to people first, then furniture, then walls. This is achieved through:
- Entities have the strongest outline weight (2px at 64x64 equivalent)
- Objects have medium outline weight (1px)
- Structure has minimal or no outline (color fills only)
- D-033 colors on entities are more saturated than any object or structure color
Gore's "simple entities, richer environments" position from Round 1 translates well here: entities are simple bold silhouettes, the OBJECT LAYER provides the environmental richness, and structure provides the spatial backbone. The "richness" is in the density and variety of placed objects, not in individual object detail.
OQ-04: Animation Ambiguity vs. Readability
The proposed resolution path is exactly right. I'm endorsing it as stated.
Clear animation for expected activities:
- Walking (directional walk cycle, 4-6 frames)
- Working at a station (seated/standing at console, periodic arm movement)
- Eating/drinking (seated at table/bar, arm-to-face cycle)
- Talking (two entities facing each other, subtle gesticulation)
- Sleeping (horizontal sprite, no movement)
These are the "life-sim substrate" animations. The player needs to read daily routine at a glance — The Sims principle. If you can't tell someone is eating vs. working vs. sleeping from across the map, the life-sim readability fails.
Ambiguous animation for investigatively relevant behaviors:
- Stopping (entity halts, but WHY? Thinking? Listening? Checking surroundings? Waiting for someone? Just resting?)
- Looking around (head rotation, but the REASON is invisible)
- Meeting someone (two entities near each other — talking? Exchanging something? Coincidence?)
- Entering an unusual space (an NPC in an area they don't usually go — are they lost? On an errand? Sneaking?)
- Handling an object (hands on something — what? Can't tell at this scale)
Gore's "legible but ambiguous" framing is correct for these. You SEE the behavior. You DON'T know the motivation. That's the information gap the player fills through investigation — the monologue, the insert data, other observation channels. If the animation told you WHY someone stopped, the observation mechanic loses its purpose.
Production implication: Expected activities need 2-3 distinct animation states each. Investigatively ambiguous behaviors need only 1-2 generic states ("idle-aware," "hands-occupied") that COULD be many things. This is cheaper to produce AND better for gameplay. Win-win.
OQ-05: Darkwood as Primary Lighting Reference
Yes. I endorse Darkwood as THE primary reference for vision cone lighting.
I didn't cite it in Round 1 — that's an oversight. Having reviewed Ozzie's detailed advocacy, Darkwood solves the exact problem we have: how does restricted LOS feel emotionally in a top-down game?
Darkwood's lighting does three things we need:
-
Vision cone as emotional gradient. Full visibility in the cone, degraded at the edges, dark beyond. Not binary (XCOM) — graduated. This maps directly to our four-state fog (visible → fog-edge → hidden → remembered).
-
Light pooling. Practical light sources (lamps, fires) create pools of safety. Between the pools: darkness. The player's vision cone interacts with these pools — you can see further into a lit area than a dark one. This gives our zone lighting gameplay weight: the bar's warm lights aren't just atmosphere, they're EXTENDED VISIBILITY.
-
The darkness has weight. Darkwood's unseen space isn't just "grey tiles" — it's oppressive, heavy, alive. We want a gentler version of this. Our darkness is calm, neutral, simply "unknown" — not actively threatening like Darkwood's. But the WEIGHT of the darkness, the sense that unseen space is substantive rather than decorative, that's exactly right.
Where we diverge from Darkwood:
- Darkwood is HORROR. Our darkness is UNCERTAINTY. Darkwood's dark wants to kill you. Our dark just doesn't inform you. Different emotional register, same visual technique.
- Darkwood's vision cone is the player's only light in hostile darkness. Ours interacts with environmental lighting — the station has its own lights. The cone reveals what's beyond the station's illumination.
- Darkwood has no daily-life warmth. We need the Darkwood vision cone system to coexist with warm, well-lit social spaces where the cone barely matters because everything is already visible.
Reference hierarchy for lighting:
- Darkwood — Vision cone mechanics, light pooling, darkness-as-weight
- Blade Runner 2049 / Roger Deakins — Color temperature as emotional language
- Edward Hopper (Nighthawks) — Warm interior light surrounded by unknowable dark (Gore's reference, and it's our central image)
These three aren't competing — they're describing different aspects of the same lighting system. Darkwood for the mechanic. Deakins for the color language. Hopper for the emotional composition.
Proposed Prompt for Mood Board Image #8
"World as Collection of 1x1 Placed Items"
The image should capture: a top-down view of a space station interior where the world is VISIBLY assembled from discrete grid-aligned objects. Not a painted scene — an arrangement.
Prompt:
Top-down view of a small section of a space station interior on a visible tile grid. The floor is composed of uniform grey-blue tiles. Against the walls: individual objects placed on the grid — a desk with a terminal (2x1 tiles), a chair (1x1), a storage crate (1x1), a wall-mounted console (1x1 on wall edge), a small table with two chairs (2x2 arrangement). Each object is a distinct, bold silhouette with clean edges, clearly separate from its neighbors. The style is clean 2D illustration, muted palette, flat lighting on objects (no baked shadows — the objects are shape templates). One section shows a transition: grey institutional floor tiles meet warmer-toned floor tiles where someone has renovated a corner into a break area with a coffee setup and a mismatched chair. A doorway gap in the wall shows a corridor beyond. Two small human figures (24x32 pixel scale relative to 64x64 tiles) stand in the space — one at the desk, one walking through the doorway. The figures are bold silhouettes, clearly smaller than the tiles they occupy. The overall impression: a functional space assembled from a vocabulary of objects, like Rimworld's placement system but with the muted, restrained aesthetic of a well-maintained space station. No neon. No grime. No pixel art. Clean, contemporary 2D illustration.
What this image should prove: That our art direction (clean 2D, bold silhouettes, muted palette, lighting-driven atmosphere) works when the entire world is composed of individually placed 1x1 objects on a grid. The Rimworld assembly method plus the mood we've converged on.
Summary of Positions
| OQ | Resolution | Status |
|---|---|---|
| OQ-01 | Geometric data + soft bloom render = precision that feels organic | Resolved — synthesis of Araminta + Gore |
| OQ-02 | Strict zero environmental shift. Character-driven changes only (insert density, monologue, D-033 colors) | Resolved — moved to Gore/Miri position |
| OQ-03 | 64x64 tiles, 24x32 entities, three-layer hierarchy (structure > objects > entities), Rimworld object detail target | Resolved — clarified by base building input |
| OQ-04 | Clear animation for daily activities, ambiguous for investigatively relevant behaviors | Resolved — endorsing proposed path |
| OQ-05 | Darkwood endorsed as primary vision cone lighting reference alongside BR2049 color and Hopper composition | Resolved — endorsing Ozzie's advocacy |
| New: Base building | Confirms and strengthens Round 1 positions. Object vocabulary replaces painted backgrounds. Easier for Nano Banana pipeline. | Integrated |
All five open questions have clear positions. No new divergences introduced. Ready for synthesis.
Written by Araminta. Consistency matters more than beauty at this stage.