Screenshot

Port of krom (Peter Lemon)'s InterlaceFont demo (PeterLemon/SNES, PPU/Interlace/InterlaceFont). BG Mode 5 renders 512 horizontal pixels with 16×8 tiles; SETINI bit 0 adds screen interlace for 448 visible lines. This example lands the SDK's SETINI surface (videoSetInterlace/ObjInterlace/Overscan/PseudoHires, all composing through a write-only-register shadow) with a text page as the canonical hi-res payload — every glyph is 8 px wide, unrepresentable at 256.

The page is original art (gen_assets.py renders it and converts to Mode 5's paired-tile format; outputs committed so builds don't need Pillow). The 1-pixel stripe/checker bands are the acid test: at true 512 px they alternate every column — a 256-wide mode would alias them to flat gray.

SNES Concepts

BG Mode 5: 512 px, 16×8 tiles stored as 8×8 character pairs (N, N+1)
SETINI ($2133) via videoSetInterlace() — write-only shadow discipline
The Mode 5 trap: content displays through main AND sub screen — setMainScreen(LAYER_BG1); setSubScreen(LAYER_BG1); or odd columns stay blank
Interlace vertical addressing: tile texel rows map 1:1 to hi-res lines, so a full 448-line page needs 56 tile rows → a 32×64 tilemap

Understanding what you see

Mode 5's 512 columns interleave the sub screen (even columns) and main screen (odd columns) — a period CRT blended adjacent columns into smooth strokes, while a modern LCD's crisp pixel grid shows them individually:

Text looks slightly "double-struck" — those are the real alternating columns. Authentic PPU output (hi-res games like Seiken Densetsu 3 show the same fringing in emulators); a CRT melts it into clean strokes.
The white/black 1-px band reads as flat gray — 1-px columns are finer than typical window scaling can resolve. That IS the acid test: at 256 wide this band couldn't exist at all. On a CRT it shimmers.
The cyan/gold band still resolves into visible stripes — its period is 4 px (2 px per color), coarse enough to survive scaling: direct visual proof you are looking at 512 columns.
The 1-px checkerboard reads as gray — same physics, both axes.

Register fidelity vs the original

Register	krom	this port
BGMODE	$0D (Mode 5, priority, 16×8)	same (setMode(BG_MODE5, 0x08)) — verified $0D in luna
SETINI	$01	same (videoSetInterlace(1)) — verified $01
TM / TS	$01 / $01	same (setMainScreen+setSubScreen) — verified
BG1SC	map word $4000	same (bgSetMapPtr), 32×64 (krom: 64×32 — his page is wider, ours taller)
BG12NBA	tiles word $8000	same (bgSetGfxPtr) — verified

Validation notes (luna v1.9.0)

luna renders hi-res/interlace CONTENT correctly but captures at 256×224 (≈2×2 downsample of the 512×448 field pair) — native-resolution capture is luna#115. The full 448-line page displays once with no vertical wrap (the 32×64 map fix — a 32-row map only covers 256 hi-res lines in interlace), and the 1-px bands render as the correct blend grays.

How to Build

cd examples/backgrounds/mode5_hires && make
# regenerate the page (needs Pillow + numpy):
python3 gen_assets.py && make

Modules Used

console, dma, background