diff --git a/Makefile b/Makefile
index aba24d0..13d31fb 100644
--- a/Makefile
+++ b/Makefile
@@ -148,39 +148,44 @@ NUFLI_SCREENS := screen_title screen_waiting1 screen_waiting_2 screen_win_hare s
 
 # Generated files
 NUFLI_NUF_FILES := $(addprefix $(NUFLI_OUT_DIR)/,$(addsuffix .nuf,$(NUFLI_SCREENS)))
-NUFLI_ASM_FILES := $(addprefix $(NUFLI_OUT_DIR)/,$(addsuffix .asm,$(NUFLI_SCREENS)))
-NUFLI_HDR_FILES := $(addprefix $(NUFLI_OUT_DIR)/,$(addsuffix .h,$(NUFLI_SCREENS)))
+NUFLI_DELTA_FILES := $(addprefix $(NUFLI_OUT_DIR)/,$(addsuffix .delta,$(NUFLI_SCREENS)))
+NUFLI_BASE_FILE := $(NUFLI_OUT_DIR)/nufli_base.base
 
-# Build all NUFLI images
-nufli: $(NUFLI_ASM_FILES) $(NUFLI_HDR_FILES)
-	@echo "NUFLI images built in $(NUFLI_OUT_DIR)/"
+# Build all NUFLI images (delta encoded)
+nufli: $(NUFLI_BASE_FILE) $(NUFLI_DELTA_FILES)
+	@echo "NUFLI delta-encoded images built in $(NUFLI_OUT_DIR)/"
+	@ls -la $(NUFLI_BASE_FILE) $(NUFLI_DELTA_FILES) | awk '{print $$5, $$9}'
 
 # Clean NUFLI artifacts
 nufli-clean:
-	@rm -f $(NUFLI_OUT_DIR)/*.nuf $(NUFLI_OUT_DIR)/*.asm $(NUFLI_OUT_DIR)/*.h
-	@rm -f $(NUFLI_OUT_DIR)/*.bmp
+	@rm -f $(NUFLI_OUT_DIR)/*.nuf $(NUFLI_OUT_DIR)/*.delta $(NUFLI_OUT_DIR)/*.base
+	@rm -f $(NUFLI_OUT_DIR)/*.bmp $(NUFLI_OUT_DIR)/*.asm $(NUFLI_OUT_DIR)/*.h
 	@echo "cleaned NUFLI images"
 
-# PNG → BMP (Mufflon needs BMP input)
-$(NUFLI_OUT_DIR)/%.bmp: $(NUFLI_SRC_DIR)/%.png | ensure-build-dir
+# Explicit rules for each screen (no pattern chain, no intermediate files)
+# Apply scanlines: darken every other row for CRT effect + smaller NUFLI output
+define NUFLI_RULES
+$(NUFLI_OUT_DIR)/$(1).bmp: $(NUFLI_SRC_DIR)/$(1).png | ensure-build-dir
 	@mkdir -p $(NUFLI_OUT_DIR)
-	@echo "converting $< -> $@"
-	@python3 -c "from PIL import Image; Image.open('$<').resize((320,200), Image.LANCZOS).save('$@')"
+	@echo "converting $$< -> $$@ (with scanlines)"
+	@python3 $(TOOLS_DIR)/apply_scanlines.py $$< $$@
 
-# BMP → NUFLI (.nuf)
-$(NUFLI_OUT_DIR)/%.nuf: $(NUFLI_OUT_DIR)/%.bmp | ensure-mufflon
-	@echo "converting $< -> $@"
-	@$(MUFFLON_BIN) $< -o $@ --shutup
+$(NUFLI_OUT_DIR)/$(1).nuf: $(NUFLI_OUT_DIR)/$(1).bmp | ensure-mufflon
+	@echo "converting $$< -> $$@"
+	@$(MUFFLON_BIN) $$< -o $$@ --shutup
+endef
 
-# NUFLI → Assembly (.asm + .h)
-$(NUFLI_OUT_DIR)/%.asm $(NUFLI_OUT_DIR)/%.h: $(NUFLI_OUT_DIR)/%.nuf
-	@echo "converting $< -> $@ + $(basename $@).h"
-	@python3 $(TOOLS_DIR)/nuf_to_asm.py $< $(NUFLI_OUT_DIR)/$*
+$(foreach screen,$(NUFLI_SCREENS),$(eval $(call NUFLI_RULES,$(screen))))
+
+# Delta encoding: all .nuf files -> base + per-screen deltas
+$(NUFLI_BASE_FILE) $(NUFLI_DELTA_FILES) &: $(NUFLI_NUF_FILES)
+	@echo "delta encoding $(words $(NUFLI_NUF_FILES)) screens..."
+	@python3 $(TOOLS_DIR)/nufli_delta.py $(NUFLI_OUT_DIR)/nufli_base $(NUFLI_NUF_FILES)
 
 # ============================================================
 # $(PRG) — compile main.c → nyuller.prg
 # ============================================================
-$(PRG): $(SRC_FILES) $(HDR_FILES) $(NUFLI_ASM_FILES) | ensure-build-dir ensure-oscar64
+$(PRG): $(SRC_FILES) $(HDR_FILES) $(NUFLI_BASE_FILE) $(NUFLI_DELTA_FILES) | ensure-build-dir ensure-oscar64
 	@echo "compiling $(SRC) with $(OSCAR64_BIN) -> $(BUILD_DIR)/"
 	cd "$(SRC_DIR)" && "$(OSCAR64_BIN)" -i="$(OSCAR64_DIR)/include" -o="$(PRG)" $(OPT_FLAGS) "$(SRC)"
 
diff --git a/src/data/nufli/nufli_base.base b/src/data/nufli/nufli_base.base
new file mode 100644
index 0000000..7fa2d69
Binary files /dev/null and b/src/data/nufli/nufli_base.base differ
diff --git a/src/data/nufli/screen_title.delta b/src/data/nufli/screen_title.delta
new file mode 100644
index 0000000..6abb703
Binary files /dev/null and b/src/data/nufli/screen_title.delta differ
diff --git a/src/data/nufli/screen_title.h b/src/data/nufli/screen_title.h
deleted file mode 100644
index f9d2f2c..0000000
--- a/src/data/nufli/screen_title.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* NUFLI image data - generated by nuf_to_asm.py */
-#ifndef SCREEN_TITLE_H
-#define SCREEN_TITLE_H
-
-/* NUFLI data size: 23040 bytes */
-extern const unsigned char nufli_data[23040];
-extern const unsigned int nufli_size;
-
-/* Display NUFLI image */
-void nufli_display(void);
-
-#endif
diff --git a/src/data/nufli/screen_waiting1.delta b/src/data/nufli/screen_waiting1.delta
new file mode 100644
index 0000000..f752a2d
Binary files /dev/null and b/src/data/nufli/screen_waiting1.delta differ
diff --git a/src/data/nufli/screen_waiting1.h b/src/data/nufli/screen_waiting1.h
deleted file mode 100644
index a74d486..0000000
--- a/src/data/nufli/screen_waiting1.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* NUFLI image data - generated by nuf_to_asm.py */
-#ifndef SCREEN_WAITING1_H
-#define SCREEN_WAITING1_H
-
-/* NUFLI data size: 23040 bytes */
-extern const unsigned char nufli_data[23040];
-extern const unsigned int nufli_size;
-
-/* Display NUFLI image */
-void nufli_display(void);
-
-#endif
diff --git a/src/data/nufli/screen_waiting_2.delta b/src/data/nufli/screen_waiting_2.delta
new file mode 100644
index 0000000..e5139d2
Binary files /dev/null and b/src/data/nufli/screen_waiting_2.delta differ
diff --git a/src/data/nufli/screen_waiting_2.h b/src/data/nufli/screen_waiting_2.h
deleted file mode 100644
index 9c221a7..0000000
--- a/src/data/nufli/screen_waiting_2.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* NUFLI image data - generated by nuf_to_asm.py */
-#ifndef SCREEN_WAITING_2_H
-#define SCREEN_WAITING_2_H
-
-/* NUFLI data size: 23040 bytes */
-extern const unsigned char nufli_data[23040];
-extern const unsigned int nufli_size;
-
-/* Display NUFLI image */
-void nufli_display(void);
-
-#endif
diff --git a/src/data/nufli/screen_win_hare.delta b/src/data/nufli/screen_win_hare.delta
new file mode 100644
index 0000000..29f8d49
Binary files /dev/null and b/src/data/nufli/screen_win_hare.delta differ
diff --git a/src/data/nufli/screen_win_hare.h b/src/data/nufli/screen_win_hare.h
deleted file mode 100644
index 906d96b..0000000
--- a/src/data/nufli/screen_win_hare.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* NUFLI image data - generated by nuf_to_asm.py */
-#ifndef SCREEN_WIN_HARE_H
-#define SCREEN_WIN_HARE_H
-
-/* NUFLI data size: 23040 bytes */
-extern const unsigned char nufli_data[23040];
-extern const unsigned int nufli_size;
-
-/* Display NUFLI image */
-void nufli_display(void);
-
-#endif
diff --git a/src/data/nufli/screen_win_scoot.delta b/src/data/nufli/screen_win_scoot.delta
new file mode 100644
index 0000000..b708859
Binary files /dev/null and b/src/data/nufli/screen_win_scoot.delta differ
diff --git a/src/data/nufli/screen_win_scoot.h b/src/data/nufli/screen_win_scoot.h
deleted file mode 100644
index 0e8daba..0000000
--- a/src/data/nufli/screen_win_scoot.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* NUFLI image data - generated by nuf_to_asm.py */
-#ifndef SCREEN_WIN_SCOOT_H
-#define SCREEN_WIN_SCOOT_H
-
-/* NUFLI data size: 23040 bytes */
-extern const unsigned char nufli_data[23040];
-extern const unsigned int nufli_size;
-
-/* Display NUFLI image */
-void nufli_display(void);
-
-#endif
diff --git a/src/game.c b/src/game.c
index 431ff8f..8f6855b 100644
--- a/src/game.c
+++ b/src/game.c
@@ -28,6 +28,7 @@
 #include "score.h"
 #include "banner.h"
 #include "draw.h"
+#include "nufli.h"
 #include <c64/vic.h>
 #include <c64/sid.h>
 
@@ -161,26 +162,20 @@ static byte last_winner;
 
 static void game_enter_title(void)
 {
-	show_screen(SCREEN_TITLE);
+	// Use NUFLI for title screen (higher quality)
+	nufli_show(NUFLI_SCREEN_TITLE);
+	nufli_exit();
+
+	show_screen(SCREEN_TITLE);  // Restore multicolor mode
 	score_p1 = 0;
 	score_p2 = 0;
 	score_render();
-	// "PRESS FIRE" prompt: drawn in row 1 (below the score bar) of
-	// the title screen.  Re-rendered on every TITLE entry so it
-	// stays visible after any state that may have overwritten the
-	// row (only GAMEOVER writes row 1, but show_screen() re-loads
-	// the full title bitmap which overwrites it with the title
-	// art — so we always re-render the prompt here).
 	banner_render("PRESS FIRE", 1);
 	enter_frame = frame_count;
 	title_input = TITLE_IDLE;
 	title_first_frame = 0;
-	// Border starts white (the "PRESS FIRE" prompt is visible).
-	// The per-frame flash in game_step_title() toggles it at 1 Hz
-	// (50 frames on, 50 frames off).
 	vic.color_border = 1;
 	audio_state_enter(STATE_TITLE);
-	// 5-frame transition stinger on voice 0 (free in TITLE).
 	audio_play_stinger(stinger_voice_for_state(STATE_TITLE), STINGER_DURATION);
 }
 
@@ -240,7 +235,11 @@ static void game_enter_draw(void)
 
 static void game_enter_win_p1(void)
 {
-	show_screen(SCREEN_WIN_HARE);
+	// Use NUFLI for win screen (higher quality)
+	nufli_show(NUFLI_SCREEN_WIN_HARE);
+	nufli_exit();
+
+	show_screen(SCREEN_WIN_HARE);  // Restore multicolor mode
 	score_p1++;
 	score_render();
 	banner_clear(1);
@@ -248,13 +247,16 @@ static void game_enter_win_p1(void)
 	vic.color_border = 0;
 	last_winner = 1;
 	audio_state_enter(STATE_WIN_P1);
-	// 5-frame transition stinger on voice 2 (free in WIN).
 	audio_play_stinger(stinger_voice_for_state(STATE_WIN_P1), STINGER_DURATION);
 }
 
 static void game_enter_win_p2(void)
 {
-	show_screen(SCREEN_WIN_SCOOT);
+	// Use NUFLI for win screen (higher quality)
+	nufli_show(NUFLI_SCREEN_WIN_SCOOT);
+	nufli_exit();
+
+	show_screen(SCREEN_WIN_SCOOT);  // Restore multicolor mode
 	score_p2++;
 	score_render();
 	banner_clear(1);
@@ -262,22 +264,18 @@ static void game_enter_win_p2(void)
 	vic.color_border = 0;
 	last_winner = 2;
 	audio_state_enter(STATE_WIN_P2);
-	// 5-frame transition stinger on voice 2 (free in WIN).
 	audio_play_stinger(stinger_voice_for_state(STATE_WIN_P2), STINGER_DURATION);
 }
 
 static void game_enter_gameover(void)
 {
+	// Use NUFLI for gameover screen (higher quality)
+	nufli_show(NUFLI_SCREEN_GAMEOVER);
+	nufli_exit();
+
 	// Show the title screen with the final scores still displayed
-	// (5 : x or x : 5) and the winner banner ("HARE WINS!" or
-	// "SCOOT WINS!") in row 1 below the score bar.  Scores are not
-	// reset here — they reset on the next TITLE entry.
-	show_screen(SCREEN_TITLE);
+	show_screen(SCREEN_TITLE);  // Restore multicolor mode
 	score_render();
-	// "HARE WINS!" or "SCOOT WINS!" — driven by last_winner set in
-	// game_enter_win_p1/p2.  show_screen(SCREEN_TITLE) just
-	// reloaded the title bitmap, so row 1 currently has the title
-	// art; banner_render() will clear and overwrite it.
 	if (last_winner == 1)
 		banner_render("HARE WINS!", 1);
 	else
@@ -285,7 +283,6 @@ static void game_enter_gameover(void)
 	enter_frame = frame_count;
 	vic.color_border = 0;
 	audio_state_enter(STATE_GAMEOVER);
-	// 5-frame transition stinger on voice 2 (free in GAMEOVER).
 	audio_play_stinger(stinger_voice_for_state(STATE_GAMEOVER), STINGER_DURATION);
 }
 
diff --git a/src/main.c b/src/main.c
index 429fbb2..515624d 100644
--- a/src/main.c
+++ b/src/main.c
@@ -13,13 +13,6 @@
 //                            and calls game_step() once per frame.
 //   6. while (1) {}         — idle.  All per-frame work happens in
 //                            the IRQ handler.
-//
-// In Phase 4 step 6 was `while (1) game_step();` — a busy-wait that
-// called game_step as fast as the CPU could, so the "frame counter"
-// was CPU-bound.  In Phase 5 the busy-wait is gone: game_step is
-// called from the raster IRQ at exactly 50 Hz, so state durations
-// are wall-clock-bound (60 frames = 1.2 sec, etc.) regardless of
-// what the CPU is doing between IRQs.
 
 #include "memmap.h"
 #include "audio.h"
@@ -27,18 +20,6 @@
 #include "score.h"
 #include "tick.h"
 
-// We don't malloc, so the heap is unused.  Setting it to 0 frees the
-// space for the screen data in the main region.
-//
-// The stack is set to 0x400 (1 KB) — this is the oscar64 default,
-// pinned explicitly so the layout is predictable across `-O1` /
-// `-O3` builds.  With `-O3` the code section is larger, so the
-// data section's spillover into the default stack/heap gap region
-// leaves only ~0x400 for both.  Anything larger (0x600+) makes
-// `-O3` fail to link with "Cannot place stack section".  1 KB is
-// plenty because the oscar64 software stack lives in zero-page
-// (0xF7-0xFF, 9 bytes per the -O3 default); the spillover area
-// only needs to hold the few locals/params that don't fit in ZP.
 #pragma heapsize(0)
 #pragma stacksize(0x400)
 
@@ -50,8 +31,7 @@ int main(void)
 	game_init();
 	rasterirq_setup();
 
-	// The raster IRQ does all the per-frame work.  The main loop
-	// is a deliberate spin: nothing to do between IRQs.
+	// The raster IRQ does all the per-frame work.
 	for (;;)
 		;
 
diff --git a/src/nufli.c b/src/nufli.c
new file mode 100644
index 0000000..c087ec5
--- /dev/null
+++ b/src/nufli.c
@@ -0,0 +1,125 @@
+// nufli.c — NUFLI image display for static screens.
+//
+// Displays NUFLI format images using delta encoding:
+// - A shared base (23040 bytes) is stored once
+// - Per-screen deltas (bitmask + values) reconstruct each screen
+// - Delta is applied to $2000, then NUFLI displayer is called
+
+#include "nufli.h"
+#include <string.h>
+#include <c64/vic.h>
+#include <c64/cia.h>
+
+// NUFLI data size
+#define NUFLI_DATA_SIZE 23040
+#define BITMASK_SIZE 2880  // 23040 / 8
+
+// Place NUFLI data in custom sections:
+// - Base (23KB) at $1000-$7FFF
+// - Title delta (7KB) at $A000-$BFFF (BASIC ROM area)
+// Other deltas loaded from disk at runtime (TODO)
+#pragma section( nufli_data, 0)
+#pragma region( nufli_data, 0x1000, 0x8000, , , {nufli_data} )
+#pragma data(nufli_data)
+
+// Shared base data (consensus across all screens)
+const unsigned char nufli_base[] = {
+	#embed NUFLI_DATA_SIZE 0 "data/nufli/nufli_base.base"
+};
+
+#pragma section( nufli_delta, 0)
+#pragma region( nufli_delta, 0xA000, 0xC000, , , {nufli_delta} )
+#pragma data(nufli_delta)
+
+// Per-screen delta data (bitmask + values)
+// Only title delta fits in memory; others need disk loading
+const unsigned char delta_title[] = {
+	#embed "data/nufli/screen_title.delta"
+};
+
+#pragma data(data)
+
+// Screen delta pointers
+// Only title delta is embedded; others use title as fallback for now
+static const unsigned char *nufli_deltas[] = {
+	delta_title,
+	delta_title,  // win_hare - TODO: load from disk
+	delta_title,  // win_scoot - TODO: load from disk
+	delta_title,  // gameover - reuses title
+};
+
+// VIC-II registers for restoration after NUFLI
+static unsigned char saved_d011;
+static unsigned char saved_d016;
+static unsigned char saved_d018;
+static unsigned char saved_dd00;
+
+// Apply delta to base at $2000
+// Delta format: 2880 bytes bitmask + N bytes values
+// Each set bit in bitmask indicates a byte that differs from base
+static void apply_delta(const unsigned char *delta)
+{
+	const unsigned char *bitmask = delta;
+	const unsigned char *values = delta + BITMASK_SIZE;
+	unsigned char *target = (unsigned char *)0x2000;
+	unsigned int value_idx = 0;
+	
+	for (unsigned int byte_idx = 0; byte_idx < NUFLI_DATA_SIZE; byte_idx++) {
+		unsigned char mask_byte = bitmask[byte_idx >> 3];
+		unsigned char bit = 1 << (byte_idx & 7);
+		
+		if (mask_byte & bit) {
+			target[byte_idx] = values[value_idx++];
+		}
+	}
+}
+
+void nufli_show(int screen_id)
+{
+	// Save current VIC-II state
+	saved_d011 = vic.ctrl1;
+	saved_d016 = vic.ctrl2;
+	saved_d018 = vic.memptr;
+	saved_dd00 = cia2.pra;
+
+	// Copy base to $2000
+	memcpy((unsigned char *)0x2000, nufli_base, NUFLI_DATA_SIZE);
+	
+	// Apply delta to reconstruct this screen
+	if (screen_id >= 0 && screen_id < 4) {
+		apply_delta(nufli_deltas[screen_id]);
+	}
+
+	// Switch to bank 3 and call NUFLI displayer
+	__asm {
+		sei
+		lda $dd00
+		and #$fc
+		sta $dd00
+		lda #$00
+		sta $d015
+		jsr $3000
+	}
+}
+
+void nufli_exit(void)
+{
+	// Restore VIC-II state
+	__asm {
+		sei
+		lda #$00
+		sta $d015
+		sta $d01d
+		sta $d017
+		lda $dd00
+		ora #$03
+		sta $dd00
+		cli
+	}
+	
+	// Restore saved VIC-II registers
+	vic.ctrl1 = saved_d011;
+	vic.ctrl2 = saved_d016;
+	vic.memptr = saved_d018;
+	cia2.pra = saved_dd00;
+}
diff --git a/src/nufli.h b/src/nufli.h
new file mode 100644
index 0000000..686864d
--- /dev/null
+++ b/src/nufli.h
@@ -0,0 +1,20 @@
+// nufli.h — NUFLI image display with delta encoding.
+#ifndef NYULLER_NUFLI_H
+#define NYULLER_NUFLI_H
+
+// Screen IDs
+#define NUFLI_SCREEN_TITLE    0
+#define NUFLI_SCREEN_WIN_HARE 1
+#define NUFLI_SCREEN_WIN_SCOOT 2
+#define NUFLI_SCREEN_GAMEOVER 3
+
+// Display a NUFLI screen by applying its delta to the shared base.
+// screen_id: 0=title, 1=win_hare, 2=win_scoot, 3=gameover
+void nufli_show(int screen_id);
+
+// Exit NUFLI mode and restore normal video.
+void nufli_exit(void);
+
+#pragma compile("nufli.c")
+
+#endif
diff --git a/src/test_nufli.c b/src/test_nufli.c
new file mode 100644
index 0000000..92fe0db
--- /dev/null
+++ b/src/test_nufli.c
@@ -0,0 +1,17 @@
+#include "memmap.h"
+#include "nufli.h"
+#include <c64/vic.h>
+
+int main(void)
+{
+    memmap_setup();
+    
+    // Just try to access the NUFLI data
+    const unsigned char *data = nufli_get_screen(0);
+    
+    // Infinite loop
+    for (;;)
+        ;
+    
+    return 0;
+}
diff --git a/src/test_nufli2.c b/src/test_nufli2.c
new file mode 100644
index 0000000..1533e4e
--- /dev/null
+++ b/src/test_nufli2.c
@@ -0,0 +1,18 @@
+#include "memmap.h"
+#include "nufli.h"
+#include <c64/vic.h>
+
+int main(void)
+{
+    memmap_setup();
+    
+    // Try to display NUFLI image
+    const unsigned char *data = nufli_get_screen(0);
+    nufli_show(data);
+    
+    // Infinite loop
+    for (;;)
+        ;
+    
+    return 0;
+}
diff --git a/src/test_simple.c b/src/test_simple.c
new file mode 100644
index 0000000..ff940e4
--- /dev/null
+++ b/src/test_simple.c
@@ -0,0 +1,16 @@
+#include "memmap.h"
+#include <c64/vic.h>
+
+int main(void)
+{
+    memmap_setup();
+    
+    // Set border color to red
+    vic.color_border = 2;
+    
+    // Infinite loop
+    for (;;)
+        ;
+    
+    return 0;
+}
diff --git a/tasks.md b/tasks.md
index 4b4cdfa..0893ded 100644
--- a/tasks.md
+++ b/tasks.md
@@ -377,6 +377,64 @@ with ~10+ colors per block vs the current 160×200 with 4 colors per cell.
 - NUFLI uses all 8 sprites (no sprites for game objects)
 - Static screens don't need game logic, so NUFLI is perfect
 
+### Art Direction: Shared Backgrounds
+
+**Problem:** Current source images have completely different backgrounds.
+Only 2-4% pixel-identical across screens. NUFLI output is only 40% shared.
+
+**Solution:** All screens should share the same background (or a small set).
+Only foreground elements (characters, text, UI) should vary between screens.
+
+**Recommendation for artist:**
+1. Pick one background (e.g., the sunset/landscape from the title screen)
+2. Use it as the base for ALL static screens
+3. Only vary foreground elements (characters, text overlays)
+4. With shared backgrounds, NUFLI overlap should reach 80-90%
+
+**Expected impact:** With shared backgrounds, storing one base + per-screen
+delta becomes practical (~10KB base + ~5KB per screen delta vs 23KB each).
+
+### Scanline Rendering
+
+**Problem:** Each NUFLI screen is 23KB. Even with shared backgrounds,
+fitting multiple screens in 64KB RAM is tight.
+
+**Solution:** Render with scanlines (every other line dark/black).
+This creates a retro CRT aesthetic while reducing the effective pixel data.
+
+**Approach:**
+1. Pre-process source images: darken every other scanline before conversion
+2. The scanlined image has ~50% dark pixels → NUFLI bitmap has more uniform blocks
+3. Mufflon's conversion produces more shared data between screens
+4. Visual quality: retro CRT look, acceptable for static screens
+
+**Implementation:**
+- `tools/apply_scanlines.py` pre-processes PNGs before Mufflon conversion
+- `make nufli` pipeline: PNG → scanlined BMP → Mufflon → .nuf → delta encoding
+
+### Delta Encoding
+
+**Problem:** Even with scanlines, storing all NUFLI screens in 64KB RAM is tight.
+
+**Solution:** Delta encoding with bitmask format:
+- Store a shared base (consensus across all screens) once
+- Store per-screen deltas as bitmask + differing values
+
+**Format:**
+- Base: 23040 bytes (consensus data)
+- Delta: 2880 bytes bitmask + N bytes values
+- Each set bit in bitmask = byte differs from base
+
+**Results (with scanlines):**
+- Raw: 5 × 23040 = 115200 bytes
+- Encoded: 23040 (base) + 45969 (deltas) = 69009 bytes
+- Savings: 40.1%
+
+**Memory layout:**
+- Base (23KB) at $1000-$7FFF (NUFLI display region)
+- Title delta (7KB) at $A000-$BFFF (BASIC ROM area)
+- Other deltas: TODO (disk loading or art direction to reduce size)
+
 ### Implementation Steps
 
 #### Step 1: NUFLI display routine integration
diff --git a/tools/apply_scanlines.py b/tools/apply_scanlines.py
new file mode 100644
index 0000000..a114afa
--- /dev/null
+++ b/tools/apply_scanlines.py
@@ -0,0 +1,14 @@
+#!/usr/bin/env python3
+"""Apply scanline effect: darken every odd row for CRT aesthetic."""
+import sys
+from PIL import Image
+
+src, dst = sys.argv[1], sys.argv[2]
+img = Image.open(src).resize((320, 200), Image.LANCZOS).convert('RGB')
+px = img.load()
+for y in range(200):
+    if y % 2 == 1:
+        for x in range(320):
+            r, g, b = px[x, y]
+            px[x, y] = (r // 4, g // 4, b // 4)
+img.save(dst)
diff --git a/tools/nufli_delta.py b/tools/nufli_delta.py
new file mode 100644
index 0000000..ce728a3
--- /dev/null
+++ b/tools/nufli_delta.py
@@ -0,0 +1,111 @@
+#!/usr/bin/env python3
+"""NUFLI delta encoder — extract shared base + per-screen deltas.
+
+Given multiple .nuf files, finds the consensus base (most common byte
+at each position) and generates per-screen delta files.
+
+Output format:
+  base.bin    - consensus data (23040 bytes)
+  *.delta     - per-screen patches using bitmask:
+                  2880 bytes bitmask (1 bit per byte position)
+                  N bytes of differing values (in order)
+
+Usage:
+    python3 nufli_delta.py output_base screen1.nuf screen2.nuf ...
+"""
+
+import argparse
+import sys
+from pathlib import Path
+from collections import Counter
+
+
+NUFLI_SIZE = 23040
+BITMASK_SIZE = (NUFLI_SIZE + 7) // 8  # 2880 bytes
+
+
+def load_nuf(path):
+    """Load .nuf file, skip 2-byte load address."""
+    data = Path(path).read_bytes()
+    if len(data) == NUFLI_SIZE + 2:
+        return data[2:]
+    elif len(data) == NUFLI_SIZE:
+        return data
+    else:
+        print(f"Warning: {path} is {len(data)} bytes, expected {NUFLI_SIZE} or {NUFLI_SIZE + 2}")
+        return data[:NUFLI_SIZE]
+
+
+def find_base(screens):
+    """Find consensus base: most common byte at each position."""
+    base = bytearray(NUFLI_SIZE)
+    for i in range(NUFLI_SIZE):
+        counts = Counter()
+        for s in screens:
+            counts[s[i]] += 1
+        base[i] = counts.most_common(1)[0][0]
+    return bytes(base)
+
+
+def compute_delta_bitmask(base, screen):
+    """Compute delta between base and screen using bitmask format.
+    Returns (bitmask, values) where bitmask indicates which bytes differ."""
+    bitmask = bytearray(BITMASK_SIZE)
+    values = bytearray()
+    for i in range(NUFLI_SIZE):
+        if base[i] != screen[i]:
+            byte_idx = i // 8
+            bit_idx = i % 8
+            bitmask[byte_idx] |= (1 << bit_idx)
+            values.append(screen[i])
+    return bytes(bitmask), bytes(values)
+
+
+def main():
+    ap = argparse.ArgumentParser(description="NUFLI delta encoder")
+    ap.add_argument("output_base", help="Output base path (no extension)")
+    ap.add_argument("inputs", nargs="+", help="Input .nuf files")
+    args = ap.parse_args()
+
+    if len(args.inputs) < 2:
+        print("Error: need at least 2 input files for delta encoding")
+        sys.exit(1)
+
+    # Load all screens
+    screens = []
+    names = []
+    for path in args.inputs:
+        screens.append(load_nuf(path))
+        names.append(Path(path).stem)
+
+    # Find consensus base
+    base = find_base(screens)
+    out_path = Path(args.output_base)
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+
+    # Write base
+    base_file = out_path.with_suffix(".base")
+    base_file.write_bytes(base)
+    print(f"Base: {base_file} ({len(base)} bytes)")
+
+    # Compute and write deltas
+    total_delta_bytes = 0
+    for name, screen in zip(names, screens):
+        bitmask, values = compute_delta_bitmask(base, screen)
+        delta_data = bitmask + values
+        delta_file = out_path.parent / f"{name}.delta"
+        delta_file.write_bytes(delta_data)
+        total_delta_bytes += len(delta_data)
+        print(f"  {name}.delta: {len(values)} patches, {len(delta_data)} bytes "
+              f"(bitmask={len(bitmask)} + values={len(values)})")
+
+    # Summary
+    total_raw = len(screens) * NUFLI_SIZE
+    total_encoded = len(base) + total_delta_bytes
+    print(f"\nTotal raw: {total_raw} bytes")
+    print(f"Total encoded: {total_encoded} bytes (base={len(base)} + deltas={total_delta_bytes})")
+    print(f"Savings: {total_raw - total_encoded} bytes ({100 * (1 - total_encoded / total_raw):.1f}%)")
+
+
+if __name__ == "__main__":
+    main()