VTF (Valve Texture Format)

Source Engine

Source 2

The Valve Texture Format (VTF) is the proprietary texture format used by the Source engine. VTF files are generally referenced in a Material instead of being accessed directly, which allows re-use in different ways.

Tip:It is possible, to an extent, to restore DXT compressed VTF files (i.e. removing compression artifacts) by using an AI.

Limitations and Caveats

Stub

This article or section is a stub. You can help by expanding it.

Warning:Compatibility isn't trivial. See versions ↓ for details.

Note:All block-compressed VTF files (such as those which use DXT compression) must have dimensions that are a multiple of 4. Although VTF dimensions are usually restricted to powers of two, some third-party tools can import and export non-powers of two textures. If you opt to use non-power-of-two textures, expect texture filtering to break (this can be "remedied" by setting the point sampling flag on the texture).

Additionally, non-power-of-two VTFs are not read correctly by Hammer (and possibly in-game), being treated as the next largest power of two.

Clarify: Mipmaps under 4x4 are still stored as 4x4 blocks, but only part of them are used.

Risk of Confusion:Unlike texture formats such as DDS, which store mipmaps from largest to smallest, VTF files store mipmaps from smallest to largest.

Tools

Stub

This article or section is a stub. You can help by expanding it.

VTF files can be created from TGA or PFM images using the Source SDK Tool Vtex. The following third party tools can also be used to create, view, or edit VTFs.

Viewing

• Windows thumbnail handler (Windows XP+)
• VTF Explorer (Windows; can explore inside GCFs)
• gdk-pixbuf-vtf (Gnome\Nautilus\Eog)
• IrfanView plugin
• Leadworks Image Viewer ^{[dead link]}

(supports VTF)

Editing

• VTFEdit (GUI) and VTFCmd (command line)
• Photoshop Plug-in (6 and to CC 2018, 32-bit version only)
• GIMP Plug-in
• Paint.NET Plug-in ^{[dead link]}

– the official website

• Paint.NET Plug-in – archive on Wayback Machine
• 3DSMax plugin (versions 6 to 2009)
• VIDE

Conversion

• VTFVer (7.5 to 7.4)
• no_vtf
• 7.5 to 7.4 Batch Converter
• Easy VTF Version (7.0-7.5) Converter
• vtex2 (Supports v7.6)

Files

Both textures and materials are stored in subfolders of <mod_dir>/materials/, or pak01_dir.vpk/materials/.

Note:In older Source games like Team Fortress 2 and Half-Life 2, as well as some third-party games/mods, Valve and/or other developers separated the textures and materials into different VPKs. Typically, textures are located in <mod_dir>/<game>_textures.vpk while materials are in <mod_dir>/<game>_misc.vpk. Most Source 2009 games and earlier (including Source SDK Base 2006, 2007) stores textures inside vpks/depot_<appid>_dir.vpk.

Storage capabilities

The VTF image format can store either a flat texture, an environment map, or a volumetric texture. Each of these can have multiple frames.

• Textures up to 4096x4096 pixels.

Note:Textures with dimensions larger than 4096 pixels can be created, if a few conditions are met. The resulting .vtf should be below 33 MB (or 33,554,432 bytes)(not in ), the resolutions used is a powers of two and isn't exceeding 65,536 pixels in either dimension and the total amount of pixels of the image should be below 16.7 megapixels: For example, a texture the size of 512x32,768 will work, because 512 multiplied by 32,768 equals 16.7 megapixels.

• An environment map is a six-faced cube map, unless the version is less than 7.5 and the first frame in the header is -1, then it will be seven-faced.
• A volumetric texture is a texture with depth, where each frame is a "layer" which are layered in the third dimension. So a 16x16x16 volumetric texture has 16 separate 16x16 textures stacked to give depth. This format is used internally by Source, and you shouldn't have any need to actually create one yourself.
• For each frame and face, the VTF file contains both the basic original source image data (pixel map) and a series of mipmaps used for rendering the texture over varying distances. For block compressed formats such as DXT1, the source image dimensions must be a multiple of 4, since block compressed formats compress the image's data into 4x4 chunks. Although the source image may be rectangular, square mipmaps are stored more efficiently in the VTF.
• Start frame (for animations)
• Bump map scale
• A Reflectivity value for use by VRAD
• A very low resolution copy of the VTF for color sampling by the engine.

Resources

VTF 7.3 added an extensible "resource data" system. Anything can be added, but Source will recognise only the following:

• A CRC value, for detecting data corruption, usually accompanied by the 0x2 flag.
• An U/V LOD control. This is the highest mipmap which should be loaded when the game's Texture Detail setting is "High" (mat_picmip 0). An U LOD Control value of 11 will select the mipmap which is 2048 pixels (2¹¹) across at the "High" texture setting, and 4096 pixels across at the "Very High" (mat_picmip -1) texture setting.
  Note:Since users are currently only presented with one texture detail setting above High, there is little point setting this value to anything except 50% or 100% of your texture's size. Otherwise, the user would need to manually set theirmat_picmipto a lower value to see the full-res texture, like in , which requiressv_cheatsin most other titles.
• Animated particle sheet data.
• Expanded texture settings. This is a collection of 32 flags, none of which are in use by Valve. Unlike the built-in VTF flags these can be defined on a game-by-game basis.

Image data formats

The VTF image format can store image data in a variety of formats. Some formats were meant for the engine, some only as an interim format for conversions. The uncompressed formats are not lossy and the compressed (DXT) formats are.

Note:The VTF format can actually store many more formats than those listed below -- consult your engine branch's public/bitmap/imageformat.h to see exactly what it supports. However, these formats are not GUARANTEED to work and their enum values are prone to being shuffled around between engine branches.

Image data format table

Format	R	G	B	I	P	A	T	Comp	Sup	Comments
A8	0	0	0	N/A	N/A	8	8	False	True	No color (fully black), 8-bit alpha
ABGR8888	8	8	8	N/A	N/A	8	32	False	True	Uncompressed texture with 8-bit alpha
ARGB8888	8	8	8	N/A	N/A	8	32	False	True	Uncompressed texture with 8-bit alpha, similar to ABGR8888
BGR565	5	6	5	N/A	N/A	0	16	False	True	Uncompressed opaque texture, limited color depth
BGR888	8	8	8	N/A	N/A	0	24	False	True	Uncompressed opaque texture, full color depth
BGR888_BLUESCREEN	8	8	8	N/A	N/A	0	24	False	True	Same as BGR888, but blue pixels (hex color #0000ff) are rendered transparent instead.
BGRA4444	4	4	4	N/A	N/A	4	16	False	True	Uncompressed texture with alpha, half color depth
BGRA5551	5	5	5	N/A	N/A	1	16	False	True	Uncompressed texture, limited color depth, 1-bit alpha
BGRA8888	8	8	8	N/A	N/A	8	32	Either	True	Compressed HDR ↑ texture with no alpha or uncompressed SDR texture with 8-bit alpha
BGRX5551	5	5	5	N/A	N/A	1	16	False	True
BGRX8888	8	8	8	N/A	N/A	8	32	False	True	Like BGRA8888, but the alpha channel is always set to 255, making it functionally equivalent to BGR888.
DXT1 DXT1_ONEBITALPHA	N/A	N/A	N/A	N/A	4	N/A	4	True	True	Standard compression, optional 1-bit alpha (recommended for opaque). DXT1_ONEBITALPHA format does not properly work; use regular DXT1 with 1-bit alpha flag enabled instead.
DXT3	N/A	N/A	N/A	N/A	4	4	8	True	True	Standard compression, uninterpolated 4-bit Alpha
DXT5	N/A	N/A	N/A	N/A	4+4	N/A	8	True	True	Standard compression, interpolated 8-bit alpha (recommended for transparent/translucent)
I8	N/A	N/A	N/A	8	N/A	N/A	8	False	True	Luminance (Grayscale), no alpha
IA88	N/A	N/A	N/A	8	N/A	8	16	False	True	Luminance (Grayscale), 8-bit alpha
P8	N/A	N/A	N/A	N/A	8	N/A	8	False	False	256-color paletted
RGB565	5	6	5	N/A	N/A	0	16	False	True	Uncompressed texture, limited color depth, similar to BGR565. Not properly supported in all branches; prefer BGR565 instead, which always works.
RGB888	8	8	8	N/A	N/A	0	24	False	True	Uncompressed opaque texture, similar to BGR888.
RGB888_BLUESCREEN	8	8	8	N/A	N/A	0	24	False	True	Same as RGB888, but blue pixels (hex color #0000ff) are rendered transparent instead.
RGBA16161616	16	16	16	N/A	N/A	16	64	False	True	Integer HDR Format
RGBA16161616F	16	16	16	N/A	N/A	16	64	False	True	Floating Point HDR Format
RGBA8888	8	8	8	N/A	N/A	8	32	False	True	Uncompressed texture with 8-bit alpha, similar to ABGR8888
UV88	8	8	N/A	N/A	N/A	N/A	16	False	True	Uncompressed du/dv Format
UVLX8888	N/A	N/A	N/A	N/A	N/A	N/A	32	False	True
UVWQ8888	N/A	N/A	N/A	N/A	N/A	N/A	32	False	True

Note:Certain 16-bit (ex: BGR565, BGRA4444) and 24-bit (ex: BGR888, RGB888) formats are converted to 32-bit (BGRA8888 in DX9, RGBA8888 in DX11) before being uploaded to the GPU; this means their main purpose is for saving installation size.

Tip:If making a texture for Strata Source, prefer using Deflate-compressed VTFs in RGBA8888 (or BC7) format instead of using 16-bit or 24-bit formats, as Deflate offers better filesize savings without any quality loss; needing to both decompress the texture and convert to 32-bit would increase load times unnecessarily.

HDR compression

HDR textures can be stored in compressed form using any format with a non-binary alpha channel (BGRA8888 is greatly preferred and is what is used by Vtex, but DXT5 or BGRA4444 also work).

The formula to convert colors from 8 bits-per-channel back to 16 bits-per-channel integer HDR is:

RGB = RGB * (A * 16)

and for 16 bits-per-channel floating point HDR:

RGB = (RGB * (A * 16)) / 262144

Vtex can convert 32 bits-per-channel  PFM files to this format, and no_vtf can convert compressed HDR VTFs to regular 16 bits-per-channel  EXR files.

Note:Interpreting the data as HDR is done in the shader, and is currently only supported by the Sky shader. The sky shader also has an alternate "HDR compression method B", which uses multiple textures (at different exposures?^[confirm]) to create an HDR image.
HDR cubemaps are always stored as uncompressed RGBA16161616F, except in Strata Source, where they can be stored as BC6H.

Additional data formats

These are additional image formats that are found in imageformat.h of Source 2013 and Alien Swarm engine branch.

Formats found in Source 2013

Format	Red Bits	Green Bit	Blue Bits	Alpha Bits	Total Bits	Compressed	Supported	Comments
R32F	32	0	0	0	32	False	?	Single-channel 32-bit floating point
RGB323232F	32	32	32	0	48	False	?	32-Bit Floating point HDR
RGBA32323232F	32	32	32	32	64	False	?	32-Bit Floating point HDR w/alpha channel

Formats found in Alien Swarm engine branch

Clarify: How are the RGB10A2 and R16f formats "either" compressed or uncompressed?

Format	Red Bits	Green Bit	Blue Bits	Alpha Bits	Total Bits	Compressed	Supported	Comments
RG1616F	16	16	0	0	32	False	?	Dual-channel 16-bit floating point
RG3232F	32	32	0	0	64	False	?	Dual-channel 32-bit floating point
RGBX8888	8	8	8	8	32	False	?	Like RGBA8888, but alpha channel is always set to 255, making it functionally equivalent to RGB888.
RGBA1010102	10	10	10	2	32	Either	?	10 bit-per component render targets
BGRA1010102	10	10	10	2	32	Either	?	10 bit-per component render targets
R16F	16	0	0	0	16	Either	?	Single-channel 16 bit floating point

Additionally, Strata Source adds support for additional compression formats:

Formats found in Strata Source

Format	Red Bits	Green Bits	Blue Bits	Grey Bits	Palette Bits	Total Bits	Compressed	Supported	Comments
ATI1N	N/A	N/A	N/A	N/A	4	4	True	True	AKA BC4, greyscale; Identical to alpha channel of DXT5
ATI2N	N/A	N/A	N/A	N/A	4+4	8	True	True	AKA BC5, Red and Green; equivalent to two ATI1N textures
BC6H	N/A	N/A	N/A	N/A	8	8	True	True*	Block-compressed 16-bpc HDR textures; no alpha channel.
BC7	N/A	N/A	N/A	N/A	8	8	True	True	High-quality block compression; alpha channel optional Hardware decoding requires DX11; converted to RGBA8888 in DX9 mode[ Deprecated].

Note:Although Strata supports BC6H (as of 20 May 2023) and uses the format for HDR cubemaps, currently the only publicly available tool that can view BC6H VTFs is VPKEdit. No tool can currently create them.

Choosing an image format

Comparison of how DXTn compression affects different types of images (click to view)

Though the VTF image format provides support for a wide range of image data formats, there are only a handful of image data formats you are likely to use. These formats and their criteria are described below:

• BGR888

24 bits per pixel; no alpha.

Use this format for textures with no alpha channel and very fine gradients (i.e. normal maps or light halos).

• BGR565

16 bits per pixel; no alpha.

Use this format instead of BGR888 to save 33% of space with often negligible quality loss (especially for basetextures).

Recommended for lower resolution textures where DXT block compression would be noticeable (i.e. GoldSrc ports).

Note:BGR888 may still be desired for textures with fine gradients (such as LDR skyboxes or red glowing textures), as well as for improved accuracy for normal maps (BGR565 is incapable of properly representing a "flat" normal map).

• BGRA8888

32 bits per pixel; supports alpha.

Use this format for textures with an alpha channel and very fine gradients (i.e. normal maps or light halos). It can also be used to produce Very High quality textures.

Recommended for lower resolution textures where DXT block compression would be noticeable (i.e. GoldSrc ports).

• DXT1 (also known as BC1)

4 bits per pixel; limited alpha support.

Use this format for typical textures with no alpha channel.

Note:DXT1 supports 1 bit of alpha precision. However, any areas of 0 alpha will be fully black. This functionality is therefore best used for textures using $alphatest (although DXT5 can sometimes handle mipmaping of such textures more effectively, and should be used if using $allowalphatocoverage).

• DXT3 (also known as BC2)

8 bits per pixel; supports alpha.

DXT5 should almost always be used over DXT3, but DXT3 is acceptable (not necessarily better) for textures with an alpha channel with sharp gradients, such as if using $allowalphatocoverage; see File:DXTn alpha comparison.png.
DXT3 takes up the same file size as DXT5.

Bug:VRAD is unable to calculate texture shadows for textures using DXT3 compression. Use DXT5 or BGRA8888 for shadow-casting textures instead.  [todo tested in?]

• DXT5 (also known as BC3)

8 bits per pixel; supports alpha.

Use this format for typical textures with an alpha channel.

• I8

8 bits per pixel; no alpha.

Use this format for greyscale textures with no alpha channel and very fine gradients (i.e. light halos).

• IA88

16 bits per pixel; supports alpha.

Use this format for greyscale textures with an alpha channel and very fine gradients (i.e. smoke or light halos).

• A8

8 bits per pixel; alpha only.

Use this, in conjunction with $color, for single-color textures with an alpha channel, such as GoldSrc-style decals. By default, the texture's color will be pure white (#000000).

• RGBA16161616F

64 bits per pixel; supports alpha.

Use this format for HDR textures.

Tip:HDR skyboxes can also use a compressed format using BGRA8888, providing generally acceptable quality at half the file size. The compressed format ↑ is recommended for custom skyboxes packaged in multiplayer maps.

• UV88

16 bits per pixel; no alpha.

Use this format for DuDv maps.

Find technical details on the various DXT compression formats here and here.

Additional formats in Strata Source

Strata Source supports additional image formats in VTF v7.6, which provide better compression, reducing GPU VRAM requirements while retaining better image quality than DXT1 or DXT5.

• BC7

8 bits per pixel; supports alpha.

Use this format instead of DXT5 for textures with an alpha channel, as it uses the same amount of GPU memory.

Use this format instead of BGR888 or BGR565 when more detail is needed than provided by DXT1.

• ATI1N (also known as BC4)

4 bits per pixel; no alpha.

Use this format instead of I8 or DXT1 for greyscale textures. It uses the same amount of GPU memory as DXT1.

• BC6H

8 bits per pixel; no alpha.

Use this format for HDR textures. It is a quarter the size of compressed HDR ↑ in GPU memory, while being similar or better quality.

Note:All Source engine branches since the Orange Box branch unofficially support ATI1N and ATI2N. Despite this, they are not guaranteed to work correctly; ATI1N is interpreted as red instead of greyscale, and the Z axis of ATI2 normal maps may not be correctly generated depending on the user's GPU if using vanilla shaders^{[citation needed]}.

Texture flags

Tip:Most shader settings are configured as material parameters, not texture flags.

A VTF file can contain the following flags (version 7.5):

Flag	Value	Comment
Point Sampling	0x0001	Disable Bilinear filtering for "pixel art"-style texture filtering. Note:Breaks mipmapping, see $pointsamplemagfilter if using CS:GO branch or derivatives.
Trilinear Sampling	0x0002	Always use Trilinear filtering, even when set to Bilinear in video settings.
Clamp S	0x0004	Clamp S coordinates, to prevent horizontal texture wrapping.
Clamp T	0x0008	Clamp T coordinates, to prevent vertical texture wrapping.
Anisotropic Sampling	0x0010	Always use Anisotropic filtering, even when set to Bilinear or Trilinear in video settings.
Hint DXT5	0x0020	Used in skyboxes. Makes sure edges are seamless.[How?]
PWL Corrected	0x0040	Purpose unknown.
SRGB		Uses sRGB color space. Useful for High Gamuts. Deprecated in 7.5.
No Compress	0x0040	No DXT compression used (internal to vtex).
Normal Map	0x0080	Texture is a normal map.
No Mipmaps	0x0100	Load largest mipmap only. Does not delete existing mipmaps, just disables them.
No Level Of Detail	0x0200	Not affected by texture resolution settings.
No Minimum Mipmap	0x0400	If set, load mipmaps below 32x32 pixels. Warning:Required on Env Maps for custom shaders, like PBR, to work correctly. Tip:Use this on noisy diffuse/albedo, normal map, and detail textures to reduce aliasing when viewed at a distance.
Procedural	0x0800	Texture is an procedural texture (code can modify it).
One Bit Alpha	0x1000	One bit alpha channel used.
Eight Bit Alpha	0x2000	Eight bit alpha channel used. Note:Actually used by all formats with non-binary alpha channel (ex: BGRA4444, BGRA1616161616F)
Environment Map	0x4000	Texture is an environment map.
Render Target	0x8000	Texture is a render target.
Depth Render Target	0x10000	Texture is a depth render target.
No Debug Override	0x20000	Todo: Add description
Single Copy	0x40000	Todo: Add description
Pre SRGB	0x80000	SRGB correction has already been applied
One Over Mipmap Level In Alpha	0x80000	Fill the alpha channel with 1/Mipmap Level. Deprecated (Internal to VTEX?)
Premultiply Color By One Over Mipmap Level	0x100000	(Internal to VTEX?)
Normal To DuDv	0x200000	Texture is a DuDv map (internal to vtex).
Alpha Test Mipmap Generation	0x400000	(internal to VTEX)
No Depth Buffer	0x800000	Do not buffer for Video Processing, generally render distance.
Nice Filtered	0x1000000	NICE filtering was used to generate the mipmaps (internal to VTEX).
Clamp U	0x2000000	Clamp U coordinates (for volumetric textures).
Vertex Texture	0x4000000	Usable as a vertex texture
SSBump	0x8000000	Texture is a SSBump.
Border	0x20000000	Clamp to border colour on all texture coordinates

File format

The VTF image format is described as follows.

VTF layout

7.2	7.3 +
VTF Header VTF Low Resolution Image Data For Each Mipmap (Smallest to Largest) For Each Frame (First to Last) For Each Face (First to Last) For Each Z Slice (Min to Max; Varies with Mipmap) VTF High Resolution Image Data	VTF Header Resource entries VTF Low Resolution Image Data Other resource data ↑ For Each Mipmap (Smallest to Largest) For Each Frame (First to Last) For Each Face (First to Last) For Each Z Slice (Min to Max; Varies with Mipmap) VTF High Resolution Image Data

• Image section, or non-header data, starts slightly different between 7.2 and 7.3+ in most cases. For 7.2, it starts at the offset from the beginning of the file with the value of header size. For 7.3+, it starts immediately after the end of the header.

VTF enumerations

Image format

enum ImageFormat
{
	IMAGE_FORMAT_NONE = -1,
	IMAGE_FORMAT_RGBA8888 = 0,
	IMAGE_FORMAT_ABGR8888,
	IMAGE_FORMAT_RGB888,
	IMAGE_FORMAT_BGR888,
	IMAGE_FORMAT_RGB565,
	IMAGE_FORMAT_I8,
	IMAGE_FORMAT_IA88,
	IMAGE_FORMAT_P8,
	IMAGE_FORMAT_A8,
	IMAGE_FORMAT_RGB888_BLUESCREEN,
	IMAGE_FORMAT_BGR888_BLUESCREEN,
	IMAGE_FORMAT_ARGB8888,
	IMAGE_FORMAT_BGRA8888,
	IMAGE_FORMAT_DXT1,
	IMAGE_FORMAT_DXT3,
	IMAGE_FORMAT_DXT5,
	IMAGE_FORMAT_BGRX8888,
	IMAGE_FORMAT_BGR565,
	IMAGE_FORMAT_BGRX5551,
	IMAGE_FORMAT_BGRA4444,
	IMAGE_FORMAT_DXT1_ONEBITALPHA,
	IMAGE_FORMAT_BGRA5551,
	IMAGE_FORMAT_UV88,
	IMAGE_FORMAT_UVWQ8888,
	IMAGE_FORMAT_RGBA16161616F,
	IMAGE_FORMAT_RGBA16161616,
	IMAGE_FORMAT_UVLX8888
};

Texture flags

enum CompiledVtfFlags
{
	// Flags from the *.txt config file
	TEXTUREFLAGS_POINTSAMPLE = 0x00000001,
	TEXTUREFLAGS_TRILINEAR = 0x00000002,
	TEXTUREFLAGS_CLAMPS = 0x00000004,
	TEXTUREFLAGS_CLAMPT = 0x00000008,
	TEXTUREFLAGS_ANISOTROPIC = 0x00000010,
	TEXTUREFLAGS_HINT_DXT5 = 0x00000020,
	TEXTUREFLAGS_PWL_CORRECTED = 0x00000040,
	TEXTUREFLAGS_NORMAL = 0x00000080,
	TEXTUREFLAGS_NOMIP = 0x00000100,
	TEXTUREFLAGS_NOLOD = 0x00000200,
	TEXTUREFLAGS_ALL_MIPS = 0x00000400,
	TEXTUREFLAGS_PROCEDURAL = 0x00000800,

	// These are automatically generated by vtex from the texture data.
	TEXTUREFLAGS_ONEBITALPHA = 0x00001000,
	TEXTUREFLAGS_EIGHTBITALPHA = 0x00002000,

	// Newer flags from the *.txt config file
	TEXTUREFLAGS_ENVMAP = 0x00004000,
	TEXTUREFLAGS_RENDERTARGET = 0x00008000,
	TEXTUREFLAGS_DEPTHRENDERTARGET = 0x00010000,
	TEXTUREFLAGS_NODEBUGOVERRIDE = 0x00020000,
	TEXTUREFLAGS_SINGLECOPY	= 0x00040000,
	TEXTUREFLAGS_PRE_SRGB = 0x00080000,
        
        TEXTUREFLAGS_UNUSED_00100000 = 0x00100000,
	TEXTUREFLAGS_UNUSED_00200000 = 0x00200000,
	TEXTUREFLAGS_UNUSED_00400000 = 0x00400000,

	TEXTUREFLAGS_NODEPTHBUFFER = 0x00800000,

	TEXTUREFLAGS_UNUSED_01000000 = 0x01000000,

	TEXTUREFLAGS_CLAMPU = 0x02000000,
	TEXTUREFLAGS_VERTEXTEXTURE = 0x04000000,
	TEXTUREFLAGS_SSBUMP = 0x08000000,

	TEXTUREFLAGS_UNUSED_10000000 = 0x10000000,

	TEXTUREFLAGS_BORDER = 0x20000000,

	TEXTUREFLAGS_UNUSED_40000000 = 0x40000000,
	TEXTUREFLAGS_UNUSED_80000000 = 0x80000000,
};

VTF header

typedef struct tagVTFHEADER
{
    char            signature[4];       // File signature ("VTF\0"). (or as little-endian integer, 0x00465456)
    unsigned int    version[2];         // version[0].version[1] (currently 7.2).
    unsigned int    headerSize;         // Size of the header struct  (16 byte aligned; currently 80 bytes) + size of the resources dictionary (7.3+).
    unsigned short  width;              // Width of the largest mipmap in pixels. Must be a power of 2.
    unsigned short  height;             // Height of the largest mipmap in pixels. Must be a power of 2.
    unsigned int    flags;              // VTF flags.
    unsigned short  frames;             // Number of frames, if animated (1 for no animation).
    unsigned short  firstFrame;         // First frame in animation (0 based). Can be -1 in environment maps older than 7.5, meaning there are 7 faces, not 6.
    unsigned char   padding0[4];        // reflectivity padding (16 byte alignment).
    float           reflectivity[3];    // reflectivity vector.
    unsigned char   padding1[4];        // reflectivity padding (8 byte packing).
    float           bumpmapScale;       // Bumpmap scale.
    int             highResImageFormat; // High resolution image format.
    unsigned char   mipmapCount;        // Number of mipmaps.
    int             lowResImageFormat;  // Low resolution image format (Usually DXT1).
    unsigned char   lowResImageWidth;   // Low resolution image width.
    unsigned char   lowResImageHeight;  // Low resolution image height.

    // 7.2+
    unsigned short  depth;              // Depth of the largest mipmap in pixels. Must be a power of 2. Is 1 for a 2D texture.

    // 7.3+
    unsigned char   padding2[3];        // depth padding (4 byte alignment).
    unsigned int    numResources;       // Number of resources this vtf has. The max appears to be 32.

    unsigned char   padding3[8];        // Necessary on certain compilers
} VTFHEADER;

VTF Resource Entry

struct ResourceEntryInfo
{
	unsigned char	tag[3]; 		// A three-byte "tag" that identifies what this resource is.
	unsigned char	flags;			// Resource entry flags. The only known flag is 0x2, which indicates that no data chunk corresponds to this resource.
	unsigned int	offset;			// The offset of this resource's data in the file. 
};

Tags

{ '\x01', '\0', '\0' } - Low-res (thumbnail) image data.

{ '\x30', '\0', '\0' } - High-res image data.

{ '\x10', '\0', '\0' } - Animated particle sheet data.

{ 'C', 'R', 'C' } - CRC data.

{ 'L', 'O', 'D' } - Texture LOD control information.

{ 'T', 'S', 'O' } - Game-defined "extended" VTF flags.

{ 'K', 'V', 'D' } - Arbitrary KeyValues data.

VTF lo-res image data

Tightly packed low resolution image data in the format described in the header. The low resolution image data is always stored in the DXT1 compressed image format. Its dimensions are that of the largest mipmap with a width or height that does not exceed 16 pixels. i.e. for a 256x256 pixel VTF: 16x16, for a 256x64 pixel VTF: 16x4, for a 1x32 pixel VTF: 1x16, for a 4x4 pixel VTF: 4x4.

VTF hi-res image data

Tightly packed interleaved high resolution image data in the format described in the header. Common image formats include DXT1, DXT5, BGR888, BGRA8888 and UV88. All dimensions must be a multiple of 4.

Versions

Clarify: many dates are missing.

7.6

Unofficial VTF version supported by Strata Source. Not compatible with any of Valve's games, or other mods.

• Adds support for  deflate and  zstd compression of image data, allowing for reduced storage size.
• Fixes support for ATI1N and ATI2N, and adds support for BC6H and BC7.

7.5

• Incompatible with Source 2004 to Source 2013 (e.g. , , ) and Left 4 Dead engine branch-based games. Garry's Mod has support.
  Textures with VTF version 7.5 used by materials in those games need to be converted using a tool such as VTFVer. This is useful when e.g. mounting content from Portal 2 or Counter-Strike: Global Offensive into older games.
  • Common errors encountered when using VTFVer to convert Counter-Strike: Global Offensive textures to v7.4:
    • In some cases, an "Exception could not be handled" error manifests during texture conversion. It is not confirmed what causes this at the current time.
    • Additionally, some of Counter-Strike: Global Offensive envmap textures, such as materials/environment maps/metal_generic_xxx.vtf, fail to be converted properly and must be reconstructed in an older vtf version. This can be done by simple exporting as .tga and re-importing and saving as v7.4 or earlier.
• Released July 19th, 2010 as part of Alien Swarm.
• Bitwise equivalent to v7.4.
• Redefines and revises two texture flags.^[Elaborate?]
• Spheremaps now officially redundant.
• Most changes internal to the VTF creation process with VTEX, e.g. MipMap fading, Alpha decay and Xbox 360 formats.

7.4

• Requires Source 2007 or newer.
• Bitwise equivalent to v7.3.
• Addresses issues related to how gamma-correction is performed on textures for TV-output on XBOX 360 combined with hunting down OS Paged Pool Memory.
• Released October 10th, 2007 as part of The Orange Box and its Source engine branch.

7.3

• Requires Source 2007 or newer.
• Added an extensible resource orientated structure.
• Added CRC, Texture LOD Control and Sheet resources, along with backwards compatible Image and Low Resolution Image resources.
• Added several vendor specific depth-stencil formats (for internal engine use), along with normal map formats and linear uncompressed formats.
• Released September 18th, 2007 as part of the Team Fortress 2 beta.

7.2

• Added volumetric texture support.
• Released September 23rd, 2005 as a Steam engine update and supported in later versions of Source 2004.

7.1

• Added spheremap support to environment maps. (This was intended for DirectX 6 support which was later cut.)

7.0

• Initial release. (Internal release only, though some v7.0 textures made it to the published title.)

Other implementations

• the LGPL C/C++ library VTFLib
• cl-valve-texture-format — VTF reader and writer in Common Lisp.

See also

• Creating a Material
• Material
• VMT
• WAD, SPR, and BMP/TGA ( GoldSrc equivalents)
• VTEX ( Source 2 equivalent)
• DDS