Find all valid top-left positions for 2x2 tiles.

Find the optimal (x, y) position for the infill in the template. The position maximizes context inclusion while avoiding seams. Returns

Format bytes as human-readable string.

Format seconds into a human-readable time string.

Format seconds into a human-readable time string.

Create from JSON dict.

Parse a string like '(x,y)' or 'x,y' into a Point.

Get all 2x2 blocks that contain the point p. A point can be in up to 4 different 2x2 blocks: - As top-left - As top-right - As bottom-left

Get all 8 exterior neighbors of a 2x2 tile. For a 2x2 at (x,y), (x+1,y), (x,y+1), (x+1,y+1): - Top: (x, y-1), (x+1, y-1) - Bottom: (x, y+2),

Get all generated quadrant positions.

Get all items currently being processed (one per model in parallel mode).

Generate a cache key for a satellite tile.

Determine the infill context from a quadrant grid. Checks for generated quadrants adjacent to the selection. Args: grid: The quadrant gri

Get a connection to the generations database.

Get file size from Content-Length header.

Get the bounding box of all generated quadrants.

Get all quadrants that have been generated (have generation data). Returns list of (quadrant_x, quadrant_y, water_status) tuples. water_status

Get the generation config from the metadata table.

Get all CREATE INDEX statements for a table.

Get all unique quadrants from pending queue items. Returns list of (quadrant_x, quadrant_y) tuples.

Get either render or generation bytes for a quadrant.

Get the render bytes for a quadrant.

Get all 4 quadrant coordinates for a tile anchored at (anchor_x, anchor_y). Returns: List of (x, y) tuples for TL, TR, BL, BR quadrants

Get all 4 quadrant coordinates for a tile anchored at (anchor_x, anchor_y). Returns: List of (x, y) tuples for TL, TR, BL, BR quadrants

Get all WATER_EDGE quadrants that don't have water mask data yet. Returns: List of (x, y) coordinates for quadrants needing water masks.

Get the global web renderer instance, creating it if necessary. This provides a singleton instance of the WebRenderer that can be shared acros

Import a tile PNG into the database as quadrants. Args: db_path: Path to the quadrants.db file. tl: (x, y) coordinates of the top-le

Check if this region is exactly a full tile (1024x1024).

Check if adding new_type to current_types satisfies constraints.

Mark a queue item as complete.

Mark a queue item as errored.

Calculate minimum PMTiles zoom level to fit a grid of given size.

Parse a coordinate string like '10,20' or '10, 20' into a Point.

Parse a coordinate string like '(x,y)' or 'x,y' into a Point.

Parse a coordinate string like '0,0' or '(0,0)' into (x, y) tuple.

Parse a filename to extract coordinates and hash. Args: filename: Filename like "0012_0045_a1b2c3d4.png" Returns: Tuple of (x, y, has

Parse a comma-separated list of quadrant tuples. Args: s: String like "(0,1),(0,2)" or "(0,1), (0,2)" Returns: List of (x, y) coo

Apply pixelation and color quantization to an image. Args: img: Source image (will be converted to RGB). palette_img: Palette image

Apply unfake-style pixel art postprocessing to a tile. Args: img: Input PIL Image pixel_size: Native pixel size (default: 2)

Render a quadrant and save to database. This renders the tile containing the quadrant and saves all 4 quadrants. Args: conn: Database c

Render a 3D tile of NYC buildings. Args: lat, lon: Center coordinates size_meters: Size of the area to render orientation_deg: Rotat

Render a URL to PNG bytes using Playwright. This is a shared utility for rendering web pages to images with consistent Chromium configuration

Run the full generation pipeline for selected quadrants. This is a wrapper around run_generation_for_quadrants that ensures the web server is

Run the full nano banana generation pipeline for selected quadrants. This is the main entry point for generation. It: 1. Validates the quadran

Sample colors from quadrants in the database.

Upload a file to Google Cloud Storage and return its public URL. Args: local_path: Path to the local file to upload bucket_name: Name of

Create an ASCII visualization of the grid. Legend: G = Generated . = Empty S = Selected (highlighted)

Wait for the server to be ready by making HTTP requests. Args: port: Port to check timeout: Maximum time to wait in seconds interval

Apply a mask to ensure only the infill region has render pixels. For quadrants that partially overlap the infill region, we need to comp

Check that all generated neighbors can be included in template. Uses the generic InfillPositioner to validate that a valid position exists

Create depth-3+ plan for horizontal strip.

Create depth-3+ plan for vertical strip.

Draw a red border around the infill region.

Ensure the web server is still running, restart if needed.

Generate steps for a 3-column vertical chunk. x_cols should be ordered from closest to farthest from the generation edge. The 2x2 tiles go in

Generate steps for a 3-row horizontal chunk. y_rows should be ordered from closest to farthest from the generation edge. The 2x2 tiles go in t

Internal helper to avoid code duplication

Check if a set of positions forms a contiguous region.

Render a tile and return all 4 quadrants as PNG bytes. Returns a dict mapping (dx, dy) offset to PNG bytes.

Run basic tests.

Try alternative placements that avoid missing context quadrants. When the optimal placement would include non-generated context quadrants,

Try to find a placement that avoids missing context quadrants, even if it might create seams with generated neighbors. This is a last re

Update status via callback if available.

Validate that a placement doesn't create seams. A seam would occur if the infill region touches the template edge AND there are generate

Return all points within the bounding box.

Return all points within the strip bounds.

Apply a bounds mask to an image, blacking out pixels outside the bounds. Args: img: Input image (RGB) mask: Mask image (L mode, 255=insi

Apply a bounds mask to an image.

Apply perspective correction to satellite imagery. Args: image_array: numpy array of the image transform_params: dict with scale_x, scal

Assemble tiles into a single image using pyvips. This uses pyvips's arrayjoin to efficiently combine tiles without loading all images in

Assemble all tiles into a single pyvips image. Uses pyvips.arrayjoin for efficient memory handling. Fast path when no postprocessing: lo

Assign rectangle types to achieve ~20% distribution for each type. Args: num_images: Number of source images. variants_per_image: Nu

Build the static web app using bun/vite.

Build a template image from the specified quadrants. Builds a 1024x1024 template with all quadrants from the source layer, and a red border ar

Build a unified palette image from sampled colors. Creates a small image with the quantized palette that can be used as a reference for applyi

Build a unified palette image from sampled colors. Creates a small image with the quantized palette that can be used as a reference for applyi

Build a unified palette image from sampled colors.

Calculate the lat/lng center point of the bounding box. Args: config: Generation config dictionary top_left: (x, y) quadrant coordin

Calculate context quadrants lazily at execution time. This determines which adjacent quadrants have existing generations that can provide cont

Calculate context quadrants lazily at execution time. This determines which adjacent quadrants have existing generations that can provide cont

Calculate approximate distance between two points in meters using Haversine formula.

Calculate the new lat/lon center after shifting the view by shift_x_px and shift_y_px. Logic derived from whitebox.py camera setup: - Isometri

Calculate padded grid dimensions that are multiples of 2^max_zoom_level. This ensures the grid divides evenly at all zoom levels. Args:

Calculate padded grid dimensions that are multiples of 2^max_zoom_level.

Calculate the anchor coordinates (bottom-right corner) for all 4 quadrants of a tile. The tile center is at (tile_lat, tile_lon). Each quadrant'