diff --git a/internals/Cargo.toml b/internals/Cargo.toml
index f23271f6..a5b9d7ab 100644
--- a/internals/Cargo.toml
+++ b/internals/Cargo.toml
@@ -14,7 +14,8 @@ edition = "2018"
 # Please don't forget to add relevant features to docs.rs below.
 [features]
 default = []
-std = []
+std = ["alloc"]
+alloc = []
 
 [package.metadata.docs.rs]
 features = ["std"]
diff --git a/internals/src/hex/buf_encoder.rs b/internals/src/hex/buf_encoder.rs
new file mode 100644
index 00000000..08428bf6
--- /dev/null
+++ b/internals/src/hex/buf_encoder.rs
@@ -0,0 +1,250 @@
+//! Implements a buffered encoder.
+//!
+//! The main type of this module is [`BufEncoder`] which provides buffered hex encoding. Such is
+//! faster than the usual `write!(f, "{02x}", b)?` in a for loop because it reduces dynamic
+//! dispatch and decreases the number of allocations if a `String` is being created.
+
+pub use out_bytes::OutBytes;
+
+use super::Case;
+
+/// Implements `OutBytes`
+///
+/// This prevents the rest of the crate from accessing the field of `OutBytes`.
+mod out_bytes {
+    /// A byte buffer that can only be written-into.
+    ///
+    /// You shouldn't concern yourself with this, just call `BufEncoder::new` with your array.
+    ///
+    /// This prepares the API for potential future support of `[MaybeUninit<u8>]`. We don't want to use
+    /// `unsafe` until it's proven to be needed but if it does we have an easy, compatible upgrade
+    /// option.
+    ///
+    /// We also don't bother with unsized type because the immutable version is useless and this avoids
+    /// `unsafe` while we don't want/need it.
+    pub struct OutBytes<'a>(&'a mut [u8]);
+
+    impl<'a> OutBytes<'a> {
+        /// Returns the first `len` bytes as initialized.
+        ///
+        /// Not `unsafe` because we don't use `unsafe` (yet).
+        ///
+        /// ## Panics
+        ///
+        /// The method panics if `len` is out of bounds.
+        #[cfg_attr(rust_v_1_46, track_caller)]
+        pub(crate) fn assume_init(&self, len: usize) -> &[u8] {
+            &self.0[..len]
+        }
+
+        /// Writes given bytes into the buffer.
+        ///
+        /// ## Panics
+        ///
+        /// The method panics if pos is out of bounds or `bytes` don't fit into the buffer.
+        #[cfg_attr(rust_v_1_46, track_caller)]
+        pub(crate) fn write(&mut self, pos: usize, bytes: &[u8]) {
+            self.0[pos..(pos + bytes.len())].copy_from_slice(bytes);
+        }
+
+        /// Returns the length of the buffer.
+        pub(crate) fn len(&self) -> usize {
+            self.0.len()
+        }
+    }
+
+    macro_rules! impl_from_array {
+        ($($len:expr),* $(,)?) => {
+            $(
+                impl<'a> From<&'a mut [u8; $len]> for OutBytes<'a> {
+                    fn from(value: &'a mut [u8; $len]) -> Self {
+                        OutBytes(value)
+                    }
+                }
+            )*
+        }
+    }
+
+    // As a sanity check we only provide conversions for even, non-empty arrays.
+    // Weird lengths 66 and 130 are provided for serialized public keys.
+    impl_from_array!(2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 64, 66, 128, 130, 256, 512, 1024, 2048, 4096, 8192);
+}
+
+/// Hex-encodes bytes into the provided buffer.
+///
+/// This is an important building block for fast hex-encoding. Because string writing tools
+/// provided by `core::fmt` involve dynamic dispatch and don't allow reserving capacity in strings
+/// buffering the hex and then formatting it is significantly faster.
+pub struct BufEncoder<'a> {
+    buf: OutBytes<'a>,
+    pos: usize,
+}
+
+impl<'a> BufEncoder<'a> {
+    /// Creates an empty `BufEncoder`.
+    ///
+    /// This is usually used with uninitialized (zeroed) byte array allocated on stack.
+    /// This can only be constructed with an even-length, non-empty array.
+    #[inline]
+    pub fn new<T: Into<OutBytes<'a>>>(buf: T) -> Self {
+        let buf = buf.into();
+        BufEncoder {
+            buf,
+            pos: 0,
+        }
+    }
+
+    /// Encodes `byte` as hex in given `case` and appends it to the buffer.
+    ///
+    /// ## Panics
+    ///
+    /// The method panics if the buffer is full.
+    #[inline]
+    #[cfg_attr(rust_v_1_46, track_caller)]
+    pub fn put_byte(&mut self, byte: u8, case: Case) {
+        self.buf.write(self.pos, &super::byte_to_hex(byte, case.table()));
+        self.pos += 2;
+    }
+
+    /// Encodes `bytes` as hex in given `case` and appends them to the buffer.
+    ///
+    /// ## Panics
+    ///
+    /// The method panics if the bytes wouldn't fit the buffer.
+    #[inline]
+    #[cfg_attr(rust_v_1_46, track_caller)]
+    pub fn put_bytes(&mut self, bytes: &[u8], case: Case) {
+        // Panic if the result wouldn't fit address space to not waste time and give the optimizer
+        // more opportunities.
+        let double_len = bytes.len().checked_mul(2).expect("overflow");
+        assert!(double_len <= self.buf.len() - self.pos);
+        for byte in bytes {
+            self.put_byte(*byte, case);
+        }
+    }
+
+    /// Returns true if no more bytes can be written into the buffer.
+    #[inline]
+    pub fn is_full(&self) -> bool {
+        self.pos == self.buf.len()
+    }
+
+    /// Returns the written bytes as a hex `str`.
+    #[inline]
+    pub fn as_str(&self) -> &str {
+        core::str::from_utf8(self.buf.assume_init(self.pos)).expect("we only write ASCII")
+    }
+
+    /// Resets the buffer to become empty.
+    #[inline]
+    pub fn clear(&mut self) {
+        self.pos = 0;
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn empty() {
+        let mut buf = [0u8; 2];
+        let encoder = BufEncoder::new(&mut buf);
+        assert_eq!(encoder.as_str(), "");
+        assert!(!encoder.is_full());
+    }
+
+    #[test]
+    fn single_byte_exact_buf() {
+        let mut buf = [0u8; 2];
+        let mut encoder = BufEncoder::new(&mut buf);
+        encoder.put_byte(42, Case::Lower);
+        assert_eq!(encoder.as_str(), "2a");
+        assert!(encoder.is_full());
+        encoder.clear();
+        assert!(!encoder.is_full());
+        encoder.put_byte(42, Case::Upper);
+        assert_eq!(encoder.as_str(), "2A");
+        assert!(encoder.is_full());
+    }
+
+    #[test]
+    fn single_byte_oversized_buf() {
+        let mut buf = [0u8; 4];
+        let mut encoder = BufEncoder::new(&mut buf);
+        encoder.put_byte(42, Case::Lower);
+        assert_eq!(encoder.as_str(), "2a");
+        assert!(!encoder.is_full());
+        encoder.clear();
+        encoder.put_byte(42, Case::Upper);
+        assert_eq!(encoder.as_str(), "2A");
+        assert!(!encoder.is_full());
+    }
+
+    #[test]
+    fn two_bytes() {
+        let mut buf = [0u8; 4];
+        let mut encoder = BufEncoder::new(&mut buf);
+        encoder.put_byte(42, Case::Lower);
+        encoder.put_byte(255, Case::Lower);
+        assert_eq!(encoder.as_str(), "2aff");
+        assert!(encoder.is_full());
+        encoder.clear();
+        assert!(!encoder.is_full());
+        encoder.put_byte(42, Case::Upper);
+        encoder.put_byte(255, Case::Upper);
+        assert_eq!(encoder.as_str(), "2AFF");
+        assert!(encoder.is_full());
+    }
+
+    #[test]
+    fn same_as_fmt() {
+        use core::fmt::{self, Write};
+
+        struct Writer {
+            buf: [u8; 2],
+            pos: usize,
+        }
+
+        impl Writer {
+            fn as_str(&self) -> &str {
+                core::str::from_utf8(&self.buf[..self.pos]).unwrap()
+            }
+        }
+
+        impl Write for Writer {
+            fn write_str(&mut self, s: &str) -> fmt::Result {
+                assert!(self.pos <= 2);
+                if s.len() > 2 - self.pos {
+                    Err(fmt::Error)
+                } else {
+                    self.buf[self.pos..(self.pos + s.len())].copy_from_slice(s.as_bytes());
+                    self.pos += s.len();
+                    Ok(())
+                }
+            }
+        }
+
+        let mut writer = Writer {
+            buf: [0u8; 2],
+            pos: 0,
+        };
+        let mut buf = [0u8; 2];
+        let mut encoder = BufEncoder::new(&mut buf);
+
+        for i in 0..=255 {
+            write!(writer, "{:02x}", i).unwrap();
+            encoder.put_byte(i, Case::Lower);
+            assert_eq!(encoder.as_str(), writer.as_str());
+            writer.pos = 0;
+            encoder.clear();
+        }
+        for i in 0..=255 {
+            write!(writer, "{:02X}", i).unwrap();
+            encoder.put_byte(i, Case::Upper);
+            assert_eq!(encoder.as_str(), writer.as_str());
+            writer.pos = 0;
+            encoder.clear();
+        }
+    }
+}
diff --git a/internals/src/hex/display.rs b/internals/src/hex/display.rs
new file mode 100644
index 00000000..97549c7b
--- /dev/null
+++ b/internals/src/hex/display.rs
@@ -0,0 +1,259 @@
+//! Helpers for displaying bytes as hex strings.
+//!
+//! This module provides a trait for displaying things as hex as well as an implementation for
+//! `&[u8]`.
+
+use core::fmt;
+#[cfg(feature = "alloc")]
+use crate::prelude::*;
+use super::buf_encoder::{BufEncoder, OutBytes};
+use super::Case;
+
+/// Extension trait for types that can be displayed as hex.
+/// Types that have a single, obvious text representation being hex should **not** implement this
+/// trait and simply implement `Display` instead.
+///
+/// This trait should be generally implemented for references only. We would prefer to use GAT but
+/// that is beyond our MSRV. As a lint we require the `IsRef` trait which is implemented for all
+/// references.
+pub trait DisplayHex: Copy + sealed::IsRef {
+    /// The type providing [`fmt::Display`] implementation.
+    ///
+    /// This is usually a wrapper type holding a reference to `Self`.
+    type Display: fmt::Display;
+
+    /// Display `Self` as a continuous sequence of ASCII hex chars.
+    fn display_hex(self, case: Case) -> Self::Display;
+
+    /// Shorthand for `display_hex(Case::Lower)`.
+    ///
+    /// Avoids the requirement to import the `Case` type.
+    fn display_lower_hex(self) -> Self::Display {
+        self.display_hex(Case::Lower)
+    }
+
+    /// Shorthand for `display_hex(Case::Upper)`.
+    ///
+    /// Avoids the requirement to import the `Case` type.
+    fn display_upper_hex(self) -> Self::Display {
+        self.display_hex(Case::Upper)
+    }
+
+    /// Create a lower-hex-encoded string.
+    ///
+    /// A shorthand for `to_hex_string(Case::Lower)`, so that `Case` doesn't need to be imported.
+    ///
+    /// This may be faster than `.display_hex().to_string()` because it uses `reserve_suggestion`.
+    #[cfg(feature = "alloc")]
+    fn to_lower_hex_string(self) -> String {
+        self.to_hex_string(Case::Lower)
+    }
+
+    /// Create an upper-hex-encoded string.
+    ///
+    /// A shorthand for `to_hex_string(Case::Upper)`, so that `Case` doesn't need to be imported.
+    ///
+    /// This may be faster than `.display_hex().to_string()` because it uses `reserve_suggestion`.
+    #[cfg(feature = "alloc")]
+    fn to_upper_hex_string(self) -> String {
+        self.to_hex_string(Case::Upper)
+    }
+
+    /// Create a hex-encoded string.
+    ///
+    /// This may be faster than `.display_hex().to_string()` because it uses `reserve_suggestion`.
+    #[cfg(feature = "alloc")]
+    fn to_hex_string(self, case: Case) -> String {
+        let mut string = String::new();
+        self.append_hex_to_string(case, &mut string);
+        string
+    }
+
+    /// Appends hex-encoded content to an existing `String`.
+    ///
+    /// This may be faster than `write!(string, "{}", self.display_hex())` because it uses
+    /// `reserve_sugggestion`.
+    #[cfg(feature = "alloc")]
+    fn append_hex_to_string(self, case: Case, string: &mut String) {
+        use fmt::Write;
+
+        string.reserve(self.hex_reserve_suggestion());
+        write!(string, "{}", self.display_hex(case)).unwrap_or_else(|_| {
+            let name = core::any::type_name::<Self::Display>();
+            // We don't expect `std` to ever be buggy, so the bug is most likely in the `Display`
+            // impl of `Self::Display`.
+            panic!("The implementation of Display for {} returned an error when it shouldn't", name)
+        })
+    }
+
+    /// Hints how much bytes to reserve when creating a `String`.
+    ///
+    /// Implementors that know the number of produced bytes upfront should override this.
+    /// Defaults to 0.
+    ///
+    // We prefix the name with `hex_` to avoid potential collision with other methods.
+    fn hex_reserve_suggestion(self) -> usize {
+        0
+    }
+}
+
+mod sealed {
+    /// Trait marking a shared reference.
+    pub trait IsRef: Copy {
+    }
+
+    impl<T: ?Sized> IsRef for &'_ T {
+    }
+}
+
+impl<'a> DisplayHex for &'a [u8] {
+    type Display = DisplayByteSlice<'a>;
+
+    #[inline]
+    fn display_hex(self, case: Case) -> Self::Display {
+        DisplayByteSlice {
+            bytes: self,
+            case,
+        }
+    }
+
+    #[inline]
+    fn hex_reserve_suggestion(self) -> usize {
+        // Since the string wouldn't fit into address space if this overflows (actually even for
+        // smaller amounts) it's better to panic right away. It should also give the optimizer
+        // better opportunities.
+        self.len().checked_mul(2).expect("the string wouldn't fit into address space")
+    }
+}
+
+/// Displays byte slice as hex.
+///
+/// Created by [`<&[u8] as DisplayHex>::display_hex`](DisplayHex::display_hex).
+pub struct DisplayByteSlice<'a> {
+    bytes: &'a [u8],
+    case: Case,
+}
+
+impl<'a> fmt::Display for DisplayByteSlice<'a> {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        let mut buf = [0u8; 1024];
+        let mut encoder = super::BufEncoder::new(&mut buf);
+
+        let mut chunks = self.bytes.chunks_exact(512);
+        for chunk in &mut chunks {
+            encoder.put_bytes(chunk, self.case);
+            f.write_str(encoder.as_str())?;
+            encoder.clear();
+        }
+        encoder.put_bytes(chunks.remainder(), self.case);
+        f.write_str(encoder.as_str())
+    }
+}
+
+/// Format known-length array as hex.
+///
+/// This supports all formatting options of formatter and may be faster than calling
+/// `display_as_hex()` on an arbitrary `&[u8]`. Note that the implementation intentionally keeps
+/// leading zeros even when not requested. This is designed to display values such as hashes and
+/// keys and removing leading zeros would be confusing.
+///
+/// ## Parameters
+///
+/// * `$formatter` - a [`fmt::Formatter`].
+/// * `$len` known length of `$bytes`, must be a const expression.
+/// * `$bytes` - bytes to be encoded, most likely a reference to an array.
+/// * `$case` - value of type [`Case`] determining whether to format as lower or upper case.
+///
+/// ## Panics
+///
+/// This macro panics if `$len` is not equal to `$bytes.len()`. It also fails to compile if `$len`
+/// is more than half of `usize::MAX`.
+#[macro_export]
+macro_rules! fmt_hex_exact {
+    ($formatter:expr, $len:expr, $bytes:expr, $case:expr) => {
+        {
+            // statically check $len
+            #[allow(deprecated)]
+            const _: () = [()][($len > usize::max_value() / 2) as usize];
+            assert_eq!($bytes.len(), $len);
+            let mut buf = [0u8; $len * 2];
+            $crate::hex::display::fmt_hex_exact_fn($formatter, (&mut buf).into(), $bytes, $case)
+        }
+    }
+}
+
+// Implementation detail of `write_hex_exact` macro to de-duplicate the code
+#[doc(hidden)]
+#[inline]
+pub fn fmt_hex_exact_fn(f: &mut fmt::Formatter, buf: OutBytes<'_>, bytes: &[u8], case: Case) -> fmt::Result {
+    let mut encoder = BufEncoder::new(buf);
+    encoder.put_bytes(bytes, case);
+    f.pad_integral(true, "0x", encoder.as_str())
+}
+
+#[cfg(test)]
+mod tests {
+    #[cfg(feature = "alloc")]
+    use super::*;
+
+    #[cfg(feature = "alloc")]
+    mod alloc {
+        use super::*;
+
+        fn check_encoding(bytes: &[u8]) {
+            use core::fmt::Write;
+
+            let s1 = bytes.to_lower_hex_string();
+            let mut s2 = String::with_capacity(bytes.len() * 2);
+            for b in bytes {
+                write!(s2, "{:02x}", b).unwrap();
+            }
+            assert_eq!(s1, s2);
+        }
+
+        #[test]
+        fn empty() {
+            check_encoding(b"");
+        }
+
+        #[test]
+        fn single() {
+            check_encoding(b"*");
+        }
+
+        #[test]
+        fn two() {
+            check_encoding(b"*x");
+        }
+
+        #[test]
+        fn just_below_boundary() {
+            check_encoding(&[42; 512]);
+        }
+
+        #[test]
+        fn just_above_boundary() {
+            check_encoding(&[42; 513]);
+        }
+
+        #[test]
+        fn just_above_double_boundary() {
+            check_encoding(&[42; 1025]);
+        }
+
+        #[test]
+        fn fmt_exact_macro() {
+            use crate::alloc::string::ToString;
+
+            struct Dummy([u8; 32]);
+
+            impl fmt::Display for Dummy {
+                fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+                    fmt_hex_exact!(f, 32, &self.0, Case::Lower)
+                }
+            }
+
+            assert_eq!(Dummy([42; 32]).to_string(), "2a".repeat(32));
+        }
+    }
+}
diff --git a/internals/src/hex/mod.rs b/internals/src/hex/mod.rs
new file mode 100644
index 00000000..c4e9bbb1
--- /dev/null
+++ b/internals/src/hex/mod.rs
@@ -0,0 +1,53 @@
+//! Helpers for encoding bytes as hex strings.
+
+pub mod buf_encoder;
+pub mod display;
+
+pub use buf_encoder::BufEncoder;
+
+/// Reexports of extension traits.
+pub mod exts {
+    pub use super::display::DisplayHex;
+}
+
+/// Possible case of hex.
+#[derive(Debug, Copy, Clone, Eq, PartialEq, Hash)]
+pub enum Case {
+    /// Produce lower-case chars (`[0-9a-f]`).
+    ///
+    /// This is the default.
+    Lower,
+
+    /// Produce upper-case chars (`[0-9A-F]`).
+    Upper,
+}
+
+impl Default for Case {
+    fn default() -> Self {
+        Case::Lower
+    }
+}
+
+impl Case {
+    /// Returns the encoding table.
+    ///
+    /// The returned table may only contain displayable ASCII chars.
+    #[inline]
+    pub(crate) fn table(self) -> &'static [u8; 16] {
+        static LOWER: [u8; 16] = [b'0', b'1', b'2', b'3', b'4', b'5', b'6', b'7', b'8', b'9', b'a', b'b', b'c', b'd', b'e', b'f'];
+        static UPPER: [u8; 16] = [b'0', b'1', b'2', b'3', b'4', b'5', b'6', b'7', b'8', b'9', b'A', b'B', b'C', b'D', b'E', b'F'];
+
+        match self {
+            Case::Lower => &LOWER,
+            Case::Upper => &UPPER,
+        }
+    }
+}
+
+/// Encodes single byte as two ASCII chars using the given table.
+///
+/// The function guarantees only returning values from the provided table.
+#[inline]
+pub(crate) fn byte_to_hex(byte: u8, table: &[u8; 16]) -> [u8; 2] {
+    [table[usize::from(byte.wrapping_shr(4))], table[usize::from(byte & 0x0F)]]
+}
diff --git a/internals/src/lib.rs b/internals/src/lib.rs
index bad0d8d2..63f1a775 100644
--- a/internals/src/lib.rs
+++ b/internals/src/lib.rs
@@ -7,7 +7,7 @@
 //! [rust-bitcoin](https://github.com/rust-bitcoin) ecosystem.
 //!
 
-#![cfg_attr(all(not(feature = "std"), not(test)), no_std)]
+#![no_std]
 // Experimental features we need.
 #![cfg_attr(docsrs, feature(doc_cfg))]
 // Coding conventions
@@ -21,4 +21,17 @@
 #![deny(missing_docs)]
 #![deny(unused_must_use)]
 
+#[cfg(feature = "alloc")]
+extern crate alloc;
+
+#[cfg(feature = "std")]
+extern crate std;
+
 pub mod error;
+pub mod hex;
+
+/// Mainly reexports based on features.
+pub(crate) mod prelude {
+    #[cfg(feature = "alloc")]
+    pub(crate) use alloc::string::String;
+}