diff --git a/diff/diff.go b/diff/diff.go
index 26cf4bb7..1fc35c73 100644
--- a/diff/diff.go
+++ b/diff/diff.go
@@ -467,7 +467,7 @@ func diffMappingResults(oldContent *manifest.MappingResult, newContent *manifest
 }
 
 func diffStrings(before, after string, stripTrailingCR bool) []difflib.DiffRecord {
-	return difflib.Diff(split(before, stripTrailingCR), split(after, stripTrailingCR))
+	return diffLines(split(before, stripTrailingCR), split(after, stripTrailingCR))
 }
 
 func split(value string, stripTrailingCR bool) []string {
diff --git a/diff/lcs.go b/diff/lcs.go
new file mode 100644
index 00000000..0a764a89
--- /dev/null
+++ b/diff/lcs.go
@@ -0,0 +1,181 @@
+package diff
+
+import (
+	"github.com/aryann/difflib"
+)
+
+// diffLines computes a line-level diff between two sequences using Hirschberg's
+// linear-space LCS algorithm, producing output compatible with difflib.Diff.
+//
+// Unlike the O(N*M) space used by difflib.Diff's full-matrix dynamic
+// programming approach, this implementation requires only O(N+M) space,
+// which prevents excessive memory consumption when diffing large manifests.
+//
+// See https://github.com/databus23/helm-diff/issues/996
+func diffLines(seq1, seq2 []string) []difflib.DiffRecord {
+	start, end := numEqualStartAndEndElements(seq1, seq2)
+
+	var result []difflib.DiffRecord
+
+	for i := 0; i < start; i++ {
+		result = append(result, difflib.DiffRecord{Payload: seq1[i], Delta: difflib.Common})
+	}
+
+	mid1 := seq1[start : len(seq1)-end]
+	mid2 := seq2[start : len(seq2)-end]
+	result = append(result, hirschbergDiff(mid1, mid2)...)
+
+	for i := len(seq1) - end; i < len(seq1); i++ {
+		result = append(result, difflib.DiffRecord{Payload: seq1[i], Delta: difflib.Common})
+	}
+
+	return result
+}
+
+func numEqualStartAndEndElements(seq1, seq2 []string) (start, end int) {
+	for start < len(seq1) && start < len(seq2) && seq1[start] == seq2[start] {
+		start++
+	}
+	i, j := len(seq1)-1, len(seq2)-1
+	for i > start && j > start && seq1[i] == seq2[j] {
+		i--
+		j--
+		end++
+	}
+	return
+}
+
+// hirschbergDiff recursively computes the LCS-based diff in linear space.
+//
+// The algorithm splits seq1 in half, finds the optimal split point in seq2
+// using forward and backward LCS score rows, then recurses on each half.
+// At each recursion level only O(len(seq2)) extra space is used.
+func hirschbergDiff(seq1, seq2 []string) []difflib.DiffRecord {
+	n, m := len(seq1), len(seq2)
+
+	if n == 0 {
+		result := make([]difflib.DiffRecord, m)
+		for i, s := range seq2 {
+			result[i] = difflib.DiffRecord{Payload: s, Delta: difflib.RightOnly}
+		}
+		return result
+	}
+	if m == 0 {
+		result := make([]difflib.DiffRecord, n)
+		for i, s := range seq1 {
+			result[i] = difflib.DiffRecord{Payload: s, Delta: difflib.LeftOnly}
+		}
+		return result
+	}
+	if n == 1 {
+		return singleRowDiff(seq1[0], seq2)
+	}
+
+	mid := n / 2
+
+	forward := lcsLift(seq1[:mid], seq2)
+	backward := lcsLiftSuffix(seq1[mid:], seq2)
+
+	splitJ := 0
+	maxScore := -1
+	for j := 0; j <= m; j++ {
+		score := forward[j] + backward[j]
+		if score > maxScore {
+			maxScore = score
+			splitJ = j
+		}
+	}
+
+	left := hirschbergDiff(seq1[:mid], seq2[:splitJ])
+	right := hirschbergDiff(seq1[mid:], seq2[splitJ:])
+
+	return append(left, right...)
+}
+
+// singleRowDiff handles the base case where seq1 has exactly one element.
+// It matches difflib.Diff's behavior which prefers LeftOnly and matches
+// at the latest possible position in seq2.
+func singleRowDiff(elem string, seq2 []string) []difflib.DiffRecord {
+	result := make([]difflib.DiffRecord, 0, len(seq2)+1)
+
+	found := -1
+	for j := len(seq2) - 1; j >= 0; j-- {
+		if elem == seq2[j] {
+			found = j
+			break
+		}
+	}
+
+	if found >= 0 {
+		for j := 0; j < found; j++ {
+			result = append(result, difflib.DiffRecord{Payload: seq2[j], Delta: difflib.RightOnly})
+		}
+		result = append(result, difflib.DiffRecord{Payload: elem, Delta: difflib.Common})
+		for j := found + 1; j < len(seq2); j++ {
+			result = append(result, difflib.DiffRecord{Payload: seq2[j], Delta: difflib.RightOnly})
+		}
+	} else {
+		result = append(result, difflib.DiffRecord{Payload: elem, Delta: difflib.LeftOnly})
+		for _, s := range seq2 {
+			result = append(result, difflib.DiffRecord{Payload: s, Delta: difflib.RightOnly})
+		}
+	}
+
+	return result
+}
+
+// lcsLift computes the last row of the standard LCS DP matrix.
+// result[j] = LCS(a, b[:j]) for j = 0..len(b). Uses O(len(b)) space.
+func lcsLift(a, b []string) []int {
+	prev := make([]int, len(b)+1)
+	curr := make([]int, len(b)+1)
+
+	for i := 0; i < len(a); i++ {
+		ai := a[i]
+		for j := 0; j < len(b); j++ {
+			if ai == b[j] {
+				curr[j+1] = prev[j] + 1
+			} else if prev[j+1] >= curr[j] {
+				curr[j+1] = prev[j+1]
+			} else {
+				curr[j+1] = curr[j]
+			}
+		}
+		prev, curr = curr, prev
+		clearRow(curr)
+	}
+
+	return prev
+}
+
+// lcsLiftSuffix computes LCS(a, b[j:]) for each j = 0..len(b).
+// It reverses both sequences, runs the standard forward DP, then
+// maps the result back. Uses O(len(b)) space.
+func lcsLiftSuffix(a, b []string) []int {
+	ra := reverseStrings(a)
+	rb := reverseStrings(b)
+
+	// row[j] = LCS(ra, rb[:j]) = LCS(a, b[len(b)-j:])
+	row := lcsLift(ra, rb)
+
+	// backward[j] = LCS(a, b[j:]) = row[len(b)-j]
+	result := make([]int, len(b)+1)
+	for j := 0; j <= len(b); j++ {
+		result[j] = row[len(b)-j]
+	}
+	return result
+}
+
+func reverseStrings(s []string) []string {
+	result := make([]string, len(s))
+	for i, v := range s {
+		result[len(s)-1-i] = v
+	}
+	return result
+}
+
+func clearRow(row []int) {
+	for i := range row {
+		row[i] = 0
+	}
+}
diff --git a/diff/lcs_test.go b/diff/lcs_test.go
new file mode 100644
index 00000000..5c2f5166
--- /dev/null
+++ b/diff/lcs_test.go
@@ -0,0 +1,205 @@
+package diff
+
+import (
+	"fmt"
+	"math/rand"
+	"testing"
+
+	"github.com/aryann/difflib"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDiffLinesParityWithDifflib verifies that diffLines produces identical
+// output to difflib.Diff across a variety of inputs, ensuring the
+// linear-space implementation is a drop-in replacement.
+func TestDiffLinesParityWithDifflib(t *testing.T) {
+	cases := []struct {
+		name string
+		seq1 []string
+		seq2 []string
+	}{
+		{
+			name: "both empty",
+			seq1: nil,
+			seq2: nil,
+		},
+		{
+			name: "first empty",
+			seq1: nil,
+			seq2: []string{"a", "b"},
+		},
+		{
+			name: "second empty",
+			seq1: []string{"a", "b"},
+			seq2: nil,
+		},
+		{
+			name: "identical",
+			seq1: []string{"a", "b", "c"},
+			seq2: []string{"a", "b", "c"},
+		},
+		{
+			name: "completely different",
+			seq1: []string{"a", "b", "c"},
+			seq2: []string{"x", "y", "z"},
+		},
+		{
+			name: "common prefix only",
+			seq1: []string{"a", "b", "c"},
+			seq2: []string{"a", "b", "d"},
+		},
+		{
+			name: "common suffix only",
+			seq1: []string{"a", "b", "c"},
+			seq2: []string{"x", "b", "c"},
+		},
+		{
+			name: "interleaved changes",
+			seq1: []string{"line1", "line2", "line3", "line4", "line5"},
+			seq2: []string{"line1-changed", "line2", "line3-changed", "line4", "line5-changed"},
+		},
+		{
+			name: "insertions",
+			seq1: []string{"a", "c"},
+			seq2: []string{"a", "b", "c"},
+		},
+		{
+			name: "deletions",
+			seq1: []string{"a", "b", "c"},
+			seq2: []string{"a", "c"},
+		},
+		{
+			name: "repeated elements",
+			seq1: []string{"a", "a", "a", "b"},
+			seq2: []string{"a", "b", "a"},
+		},
+		{
+			name: "single element match at end",
+			seq1: []string{"x"},
+			seq2: []string{"a", "b", "x"},
+		},
+		{
+			name: "single element match at start",
+			seq1: []string{"x"},
+			seq2: []string{"x", "a", "b"},
+		},
+		{
+			name: "single element no match",
+			seq1: []string{"x"},
+			seq2: []string{"a", "b"},
+		},
+		{
+			name: "multiple matches of single element",
+			seq1: []string{"x"},
+			seq2: []string{"x", "a", "x", "b", "x"},
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			expected := difflib.Diff(tc.seq1, tc.seq2)
+			actual := diffLines(tc.seq1, tc.seq2)
+			require.Equal(t, expected, actual, "diffLines output differs from difflib.Diff")
+		})
+	}
+}
+
+// TestDiffLinesSemanticValidity verifies that diffLines produces semantically
+// valid diffs on random inputs. When multiple valid LCS paths exist,
+// diffLines may choose a different path than difflib.Diff, so we verify
+// correctness properties rather than exact output match.
+func TestDiffLinesSemanticValidity(t *testing.T) {
+	rnd := rand.New(rand.NewSource(42))
+
+	for iter := 0; iter < 1000; iter++ {
+		seq1 := generateRandomSequence(rnd, 1+rnd.Intn(30), 1+rnd.Intn(8))
+		seq2 := generateRandomSequence(rnd, 1+rnd.Intn(30), 1+rnd.Intn(8))
+
+		records := diffLines(seq1, seq2)
+
+		i1, i2 := 0, 0
+		commonCount := 0
+		for _, r := range records {
+			switch r.Delta {
+			case difflib.Common:
+				require.Less(t, i1, len(seq1), "iter %d: Common references seq1[%d] out of range", iter, i1)
+				require.Less(t, i2, len(seq2), "iter %d: Common references seq2[%d] out of range", iter, i2)
+				require.Equal(t, seq1[i1], r.Payload, "iter %d: Common payload mismatch at seq1[%d]", iter, i1)
+				require.Equal(t, seq2[i2], r.Payload, "iter %d: Common payload mismatch at seq2[%d]", iter, i2)
+				i1++
+				i2++
+				commonCount++
+			case difflib.LeftOnly:
+				require.Less(t, i1, len(seq1), "iter %d: LeftOnly references seq1[%d] out of range", iter, i1)
+				require.Equal(t, seq1[i1], r.Payload, "iter %d: LeftOnly payload mismatch at seq1[%d]", iter, i1)
+				i1++
+			case difflib.RightOnly:
+				require.Less(t, i2, len(seq2), "iter %d: RightOnly references seq2[%d] out of range", iter, i2)
+				require.Equal(t, seq2[i2], r.Payload, "iter %d: RightOnly payload mismatch at seq2[%d]", iter, i2)
+				i2++
+			}
+		}
+
+		require.Equal(t, len(seq1), i1, "iter %d: not all seq1 elements consumed", iter)
+		require.Equal(t, len(seq2), i2, "iter %d: not all seq2 elements consumed", iter)
+
+		expectedLCS := lcsLength(seq1, seq2)
+		require.Equal(t, expectedLCS, commonCount,
+			"iter %d: LCS length mismatch (expected %d, got %d)\nseq1=%v\nseq2=%v",
+			iter, expectedLCS, commonCount, seq1, seq2)
+	}
+}
+
+// TestDiffLinesLargeInput ensures the implementation handles large inputs
+// without excessive memory usage.
+func TestDiffLinesLargeInput(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping large input test in short mode")
+	}
+
+	N := 5000
+	seq1 := make([]string, N)
+	seq2 := make([]string, N)
+	for i := 0; i < N; i++ {
+		seq1[i] = fmt.Sprintf("line %d", i)
+		if i%3 == 0 {
+			seq2[i] = fmt.Sprintf("changed %d", i)
+		} else {
+			seq2[i] = seq1[i]
+		}
+	}
+
+	records := diffLines(seq1, seq2)
+	require.NotEmpty(t, records)
+}
+
+// lcsLength computes the LCS length using the standard O(N*M) DP.
+// Used only in tests to verify correctness.
+func lcsLength(seq1, seq2 []string) int {
+	prev := make([]int, len(seq2)+1)
+	curr := make([]int, len(seq2)+1)
+	for i := 0; i < len(seq1); i++ {
+		for j := 0; j < len(seq2); j++ {
+			if seq1[i] == seq2[j] {
+				curr[j+1] = prev[j] + 1
+			} else if prev[j+1] >= curr[j] {
+				curr[j+1] = prev[j+1]
+			} else {
+				curr[j+1] = curr[j]
+			}
+		}
+		prev, curr = curr, prev
+		for i := range curr {
+			curr[i] = 0
+		}
+	}
+	return prev[len(seq2)]
+}
+
+func generateRandomSequence(rnd *rand.Rand, length, alphabetSize int) []string {
+	result := make([]string, length)
+	for i := 0; i < length; i++ {
+		result[i] = fmt.Sprintf("item-%d", rnd.Intn(alphabetSize))
+	}
+	return result
+}