diff --git a/ebitenutil/text.png b/ebitenutil/text.png
index 9d94ec7d6..e4ebdbf8a 100644
Binary files a/ebitenutil/text.png and b/ebitenutil/text.png differ
diff --git a/internal/png/stdlibreader.go b/internal/png/stdlibreader.go
index adbe2bb51..6f8917735 100644
--- a/internal/png/stdlibreader.go
+++ b/internal/png/stdlibreader.go
@@ -53,6 +53,10 @@ func cbPaletted(cb int) bool {
 	return cbP1 <= cb && cb <= cbP8
 }
 
+func cbTrueColor(cb int) bool {
+	return cb == cbTC8 || cb == cbTC16
+}
+
 // Filter type, as per the PNG spec.
 const (
 	ftNone    = 0
@@ -872,7 +876,7 @@ func (d *decoder) parseIEND(length uint32) error {
 	return d.verifyChecksum()
 }
 
-func (d *decoder) parseChunk() error {
+func (d *decoder) parseChunk(configOnly bool) error {
 	// Read the length and chunk type.
 	if _, err := io.ReadFull(d.r, d.tmp[:8]); err != nil {
 		return err
@@ -900,6 +904,10 @@ func (d *decoder) parseChunk() error {
 			if d.stage != dsSeenPLTE {
 				return chunkOrderError
 			}
+		} else if cbTrueColor(d.cb) {
+			if d.stage != dsSeenIHDR && d.stage != dsSeenPLTE {
+				return chunkOrderError
+			}
 		} else if d.stage != dsSeenIHDR {
 			return chunkOrderError
 		}
@@ -917,6 +925,9 @@ func (d *decoder) parseChunk() error {
 			break
 		}
 		d.stage = dsSeenIDAT
+		if configOnly {
+			return nil
+		}
 		return d.parseIDAT(length)
 	case "IEND":
 		if d.stage != dsSeenIDAT {
@@ -976,7 +987,7 @@ func Decode(r io.Reader) (image.Image, error) {
 		return nil, err
 	}
 	for d.stage != dsSeenIEND {
-		if err := d.parseChunk(); err != nil {
+		if err := d.parseChunk(false); err != nil {
 			if err == io.EOF {
 				err = io.ErrUnexpectedEOF
 			}
@@ -999,21 +1010,26 @@ func DecodeConfig(r io.Reader) (image.Config, error) {
 		}
 		return image.Config{}, err
 	}
+
 	for {
-		if err := d.parseChunk(); err != nil {
+		if err := d.parseChunk(true); err != nil {
 			if err == io.EOF {
 				err = io.ErrUnexpectedEOF
 			}
 			return image.Config{}, err
 		}
-		paletted := cbPaletted(d.cb)
-		if d.stage == dsSeenIHDR && !paletted {
-			break
-		}
-		if d.stage == dsSeenPLTE && paletted {
-			break
+
+		if cbPaletted(d.cb) {
+			if d.stage >= dsSeentRNS {
+				break
+			}
+		} else {
+			if d.stage >= dsSeenIHDR {
+				break
+			}
 		}
 	}
+
 	var cm color.Model
 	switch d.cb {
 	case cbG1, cbG2, cbG4, cbG8:
diff --git a/internal/png/stdlibwriter.go b/internal/png/stdlibwriter.go
index a05054a96..95cd0d384 100644
--- a/internal/png/stdlibwriter.go
+++ b/internal/png/stdlibwriter.go
@@ -452,6 +452,36 @@ func (e *encoder) writeImage(w io.Writer, m image.Image, cb int, level int) erro
 			if nrgba != nil {
 				offset := (y - b.Min.Y) * nrgba.Stride
 				copy(cr[0][1:], nrgba.Pix[offset:offset+b.Dx()*4])
+			} else if rgba != nil {
+				dst := cr[0][1:]
+				src := rgba.Pix[rgba.PixOffset(b.Min.X, y):rgba.PixOffset(b.Max.X, y)]
+				for ; len(src) >= 4; dst, src = dst[4:], src[4:] {
+					d := (*[4]byte)(dst)
+					s := (*[4]byte)(src)
+					if s[3] == 0x00 {
+						d[0] = 0
+						d[1] = 0
+						d[2] = 0
+						d[3] = 0
+					} else if s[3] == 0xff {
+						copy(d[:], s[:])
+					} else {
+						// This code does the same as color.NRGBAModel.Convert(
+						// rgba.At(x, y)).(color.NRGBA) but with no extra memory
+						// allocations or interface/function call overhead.
+						//
+						// The multiplier m combines 0x101 (which converts
+						// 8-bit color to 16-bit color) and 0xffff (which, when
+						// combined with the division-by-a, converts from
+						// alpha-premultiplied to non-alpha-premultiplied).
+						const m = 0x101 * 0xffff
+						a := uint32(s[3]) * 0x101
+						d[0] = uint8((uint32(s[0]) * m / a) >> 8)
+						d[1] = uint8((uint32(s[1]) * m / a) >> 8)
+						d[2] = uint8((uint32(s[2]) * m / a) >> 8)
+						d[3] = s[3]
+					}
+				}
 			} else {
 				// Convert from image.Image (which is alpha-premultiplied) to PNG's non-alpha-premultiplied.
 				for x := b.Min.X; x < b.Max.X; x++ {
diff --git a/keys.go b/keys.go
index 02e992445..e41523122 100644
--- a/keys.go
+++ b/keys.go
@@ -24,7 +24,7 @@ import (
 )
 
 // A Key represents a keyboard key.
-// These keys represent pysical keys of US keyboard.
+// These keys represent physical keys of US keyboard.
 // For example, KeyQ represents Q key on US keyboards and ' (quote) key on Dvorak keyboards.
 type Key int