Base64 to BMP

How to convert Base64 to BMP

1. Paste the Base64 or the full data:image/bmp;base64,... URI, or open a .txt file with the string. Multi-megabyte strings are normal for bitmaps and are fine here.
2. The status line should say Detected BMP. The preview shows the image and its pixel size; Size gives the decoded file size in bytes.
3. Press Download file to save the .bmp.

Browsers decode BMP in an <img>, including 1-bit monochrome, 8-bit palette, 24-bit and most 32-bit files.

BMP file size from width and height

A BMP has a 14-byte file header, an info header (40 bytes in the common version), an optional palette and then the pixel rows. Each row is padded to a multiple of 4 bytes. For a 24-bit image without a palette:

file size = 54 + height × (width × 3, rounded up to a multiple of 4)

The 1-bit row adds an 8-byte, two-color palette. Because the numbers are fixed, a bitmap is easy to check: the file header stores the total size in bytes 2 to 5. In the hex dump of the 100×100 example they read 66 75 00 00, which is 0x7566 = 30,054 as a little-endian number. If Size shows less than that, the Base64 was truncated. BMP rows are stored bottom-up by default, so a cut-off bitmap loses the top of the picture.

What BMP Base64 looks like

Every BMP starts with the letters BM, so the Base64 starts with Qk. The third character depends on the file size and is 0, 1, 2 or 3; Qk02 and Qk1m are typical. The detector also checks the info header size at byte 14, so random data that happens to start with BM is not mistaken for a bitmap.

Headerless bitmaps (DIB)

Some sources hand you the bitmap without the BM file header. This is a DIB, the format of the Windows clipboard (CF_DIB), of images inside ICO files and of some device SDKs. Its Base64 starts with KAAAA, the encoding of the 40-byte info header size 28 00 00 00.

The tool cannot preview it, since without the file header the bytes are not a valid image file, and it reports binary data. The fix is to put the 14 missing bytes in front:

import base64, struct

def dib_to_bmp(dib: bytes) -> bytes:
    header_size, = struct.unpack_from("<I", dib, 0)
    bpp, = struct.unpack_from("<H", dib, 14)
    compression, colors = struct.unpack_from("<I", dib, 16)[0], struct.unpack_from("<I", dib, 32)[0]
    if bpp <= 8 and colors == 0:
        colors = 1 << bpp
    masks = 12 if header_size == 40 and compression == 3 else 0
    offset = 14 + header_size + masks + colors * 4
    return b"BM" + struct.pack("<IHHI", 14 + len(dib), 0, 0, offset) + dib

b64 = open("dib.txt").read()
with open("image.bmp", "wb") as f:
    f.write(dib_to_bmp(base64.b64decode(b64)))

The header needs the total file size and the offset of the pixel data, which follows the info header, the color masks of 16- and 32-bit bitfield images, and the palette. Encode the result again with BMP to Base64 if you need it as text, or check it here. DIBs from ICO files are a special case: their height is doubled to make room for a transparency mask. Base64 in Python has more on decoding binary data.

Common problems

• “This looks like PNG, not BMP”. The sender already converted the image, or the system only calls it a bitmap. Save it as PNG with Download as .png; it is smaller and every program reads it.
• The colors look swapped or the image is upside down in another program. BMP stores pixels as blue-green-red and usually bottom-up. A negative height in the header means top-down rows. Browsers handle all of this; simple custom readers sometimes do not.
• A device rejects the decoded file. Compare its SHA-256 with what the device received, then check the bit depth and header version in the hex dump. Older readers expect a 40-byte header (28 00 00 00 at offset 14) and no alpha channel.

FAQ