Cull a Dungeon Map Generator (Room Graph Pack 2026)

Pretty crypt paintings stall the moment you need a second floor. Searchers typing dungeon map generator on July 17, 2026 already know a single battle wallpaper is not a crawler — they need connected rooms, stable exits, and a walkable atlas a browser engine can load twice without redrawing doors by hand. This guide covers the 2026 production path: lock a room graph in AI Image Gen, snap tiles in Tileset Forge, label nodes in Canvas, then wire the pack into Phaser 4.2 with WizardGenie. Model credits, tool badges, and Phaser versions below are verified against the Sorceress source and Phaser release notes on July 17, 2026.

What a dungeon map generator outputs for indie room graphs

The phrase dungeon map generator (480/mo, KD 35 per DataForSEO probe-fresh-seeds-2.md verified July 16, 2026) targets solo and jam teams who need a playable crawl layout, not a campaign atlas. Sibling phrases — fantasy dungeon map, room graph pack, procgen floor plan, dungeon floor plan generator — all describe the same deliverable: rooms as nodes, corridors as edges, and a tile atlas that pathfinding can trust.

Production output is structured. Ship a graph card (room count, exit list, banned perspective), separate top-down PNG variants that obey that card, a Tileset Forge atlas for floors and walls, and a short JSON table with id, exits, and spawn keys so loot and lighting share vocabulary. Pair that pack with the DnD map generator sibling when you need tabletop battle-map grids and VTT cover icons — Scout answers “can tokens measure five-foot squares tonight?”; this dungeon map generator session answers “can Phaser load a reusable room graph?” When you also need coasts and kingdoms at campaign scale, route through the AI fantasy map generator guide after the dungeon graph locks.

Why one-shot wallpapers stall playable crawls

A dungeon in a browser RPG is a graph problem before it is an art problem. Players move room to room; encounters fire on node IDs; doors that only exist as painted shadows break collision. A single concept render almost always ships the wrong contract: isometric tilt that lies about corridor width, three mystery exits the card never listed, and no atlas you can reuse for the next wing.

Procedural generation pipelines fail the same way when the art layer ignores connectivity. You can randomize room templates forever and still ship a crawl that feels pasted if exits do not match the graph JSON. Tile-based games stay dominant in 2D RPG movement because walkable vs blocked is data — not a vibe. Treat the dungeon map generator as production data: lock nodes and edges first, batch orthographic top-downs second, reject any image that invents a room or paints a door three pixels wide.

Tabletop-first workflows still matter — the Scout sibling covers 5-foot battle maps and VTT export. Do that when you are running a session with tokens. When you are shipping a browser crawler, switch to a room graph pack so the engine owns spawn keys and the atlas owns collision.

The Sorceress dungeon map generator pipeline in four steps

Every shippable room graph pack ships four runtime pieces: graph card, tile atlas, labels, and a wired scene. In 2026 each maps to one Sorceress surface verified against the live catalog on July 17, 2026:

• Graph lock: write room count, exit list, grid scale, and “no text in image” before you spend credits.
• Layout generation: AI Image Gen drafts top-downs across models (Z-Image at 3 credits for fast comps, GPT Image 2 at 5 credits for cleaner orthographic locks, Flux 2 Pro at 6 credits with references — src/lib/models.ts verified July 17, 2026).
• Tile snap: Tileset Forge is Pro-badge (src/app/_home-v2/_data/tools.ts) for extract, align, preview, and export of walkable grids under the $49 lifetime unlock.
• Label and wire: Canvas is Free-badge for room IDs and door marks; WizardGenie scaffolds a Phaser 4.2 TilemapScene on your own API keys — Phaser v4.2.0 “Giedi” shipped June 19, 2026; patch 4.2.1 landed July 9, 2026 (verified against phaser.io and GitHub on July 17, 2026).

Pricing is a $49 lifetime unlock plus pay-as-you-go credit packs. Starter $10/1,000 credits, Creator $20/2,000, Plus $50/5,000, Studio $100/10,000. Verified against src/app/plans/page.tsx lines 51 and 56–59. New accounts ship with 100 starter credits. Keep WizardGenie coding on your own API key so graph wiring never burns Sorceress image credits.

Step 1 — lock a room graph and prompt AI Image Gen

Before opening AI Image Gen, write a one-page graph card. Dungeon layouts fail when each render invents its own connectivity. Lock:

• Nodes — “Four rooms: vestibule, west wing, east wing, boss crypt.”
• Edges — “Vestibule→west, vestibule→east, east→boss; no secret fifth exit.”
• Camera — “Top-down orthographic, 1:1, implied 20×20 grid, no isometric, no UI chrome, no text.”

Open AI Image Gen and paste the card as a reusable prompt prefix. Useful skeleton: “Top-down dungeon floor plan, orthographic, connected rooms matching [exit list], stone floors, torch light only, transparent or flat void outside walls, no labels, no perspective.” Run six to ten Z-Image comps at 3 credits each. Promote one or two winners with GPT Image 2 (5 credits) when corridor edges need tighter fidelity, or Flux 2 Pro with a reference (6 credits) when a later wing must match an existing stone language.

Collections are the audit trail. Tag winners graph-ash-l1-v2 and rejects reject-extra-exit so you never import a layout that breaks the JSON. For branching quest prose keyed to the same room IDs later, see the AI dungeon story generator sibling after the floor graph locks. Named residents belong in the DnD NPC generator or NPC generator roster posts — keep the dungeon map generator focused on space.

Step 2 — snap walkable tiles in Tileset Forge

Drop the winning top-down into Tileset Forge. The tool’s job is the unglamorous half of a dungeon map generator: detect tile edges, align AI art to a clean grid, preview seams, and export a sheet pathfinding can trust. Verified July 17, 2026: Tileset Forge is Pro-badge in tools.ts, covered by the $49 lifetime unlock — no per-tile credit burn for the alignment pass itself.

Audit the sheet like production data. Walk every floor tile for seam breaks. Confirm wall thickness stays consistent across doorways. Export a jam subset first (floors + walls + one door tile) so you are not polishing twenty decorative props before the player can move. If a tile refuses to align, re-prompt Image Gen with a stricter orthographic clause, or crop the offender in Canvas and re-run Forge on the cleaned crop. When you need seamless filler for large courtyards, pair Forge with Seamless Tile Gen for grass or stone loops — keep structural walls and doors on the Forge atlas so collision stays honest.

Document the atlas once. Write a one-line legend next to each tile ID (floor_a, wall_n, door_open) in a markdown file beside the PNG. Future-you will thank that file when WizardGenie asks which tile blocks diagonal movement. The dungeon map generator pipeline only scales when the atlas legend and the room JSON share vocabulary.

Step 3 — label rooms in Canvas and wire the graph in WizardGenie

Open the labeled pass in Canvas (Free). Paint short room IDs (VESTIBULE, WEST, BOSS), mark doors with a consistent icon, and keep labels outside sprite footprints when you can. Export a clean PNG for debug overlays and keep an unlabeled twin for the in-engine tilemap if your HUD will draw its own names. Name files for the loader: ash_crypt_l1.png, ash_crypt_atlas.png, ash_crypt_graph.json.

Open WizardGenie (desktop or web) and point the agent at the exported folder. Ask for a Phaser 4.2 TilemapScene that loads the Forge atlas, places spawn markers from the room table, and toggles a debug grid overlay. Cite the live engine version in your prompt: Phaser 4.2.0 Giedi (June 19, 2026) / 4.2.1 (July 9, 2026). Keep the Planner on a strong reasoner and the Executor on a cheap fast model so wiring stays affordable — WizardGenie coding uses your own API keys, not Sorceress image credits.

{
  "id": "ash_crypt_l1",
  "grid": { "cols": 20, "rows": 20 },
  "rooms": [
    { "id": "vestibule", "exits": ["west_wing", "east_wing"], "spawn": ["player"] },
    { "id": "boss_crypt", "exits": ["east_wing"], "spawn": ["boss"] }
  ],
  "tileset": "ash_crypt_atlas.png",
  "mapImage": "ash_crypt_l1.png"
}

Acceptance checks before you call the pack done: every exit in the JSON matches a painted door; pathfinding never walks through wall tiles; a second floor reuses the same atlas language without inventing a new exit list. Browse the wider tool surface at Sorceress Tools Guide. For session-runner prose around those rooms, the AI dungeon master guide is the tabletop sibling — keep it separate from the room graph pack.

What a dungeon map generator session costs on Sorceress in 2026

Pricing verified July 17, 2026 against src/app/plans/page.tsx (LIFETIME_PRICE = 49) and src/lib/models.ts. Sorceress base unlock is $49 one-time for the full tool suite. AI credits are pay-as-you-go: Starter $10 for 1,000 credits, Creator $20 for 2,000, Plus $50 for 5,000, Studio $100 for 10,000. A typical dungeon map generator session burns about 25–70 credits — six to ten Z-Image comps (18–30), one or two GPT Image 2 locks (5–10), optional BG Remover cleans — with Tileset Forge and Canvas under lifetime Pro/Free badges and WizardGenie coding on your own API keys.

Compare that to commissioning disconnected crypt paintings that never share an exit list, or redrawing doors every time the boss room moves. The integrated dungeon map generator path wins on iteration speed: change the graph card’s edges, regenerate two top-downs, re-snap Forge, re-label Canvas, reload Phaser — done in minutes. See plans for credit tiers, or start at AI Image Gen with the graph card open in another tab.

The category shift is simple: a dungeon map generator in 2026 is not a novelty wallpaper spinner. It is a production pipeline — locked room graphs, walkable atlases, labeled nodes, Phaser-ready JSON — that respects how crawls actually load in play. Sorceress built the pieces to run that pipeline in one account; your job is to kill every one-shot crypt that only looks finished at rest before you hit export. Your graph card is the product; the atlas is the runtime; the exit list is what makes the dungeon playable twice.