diff --git a/Cargo.toml b/Cargo.toml
index 60dc90c..cca2af2 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -9,9 +9,6 @@ edition = "2024"
 repository = "https://github.com/rust-av/v_frame"
 include = ["Cargo.toml", "README.md", "LICENSE", "src"]
 
-[dependencies]
-num-traits = "0.2.19"
-
 [features]
 padding_api = []
 
diff --git a/src/lib.rs b/src/lib.rs
index 5e26476..8ae738e 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -36,4 +36,5 @@
 pub mod chroma;
 pub mod frame;
 pub mod pixel;
+pub mod pixel_ext;
 pub mod plane;
diff --git a/src/pixel.rs b/src/pixel.rs
index f8a1c0e..51700ce 100644
--- a/src/pixel.rs
+++ b/src/pixel.rs
@@ -22,7 +22,6 @@
 //! - 8-bit frames must use `u8`
 //! - 9-16 bit frames must use `u16`
 
-use num_traits::PrimInt;
 use std::fmt::Debug;
 
 mod private {
@@ -56,14 +55,38 @@ mod private {
 /// i.e. using [`std::mem::zeroed`] must __not__ cause undefined behavior for
 /// implementing types.
 pub unsafe trait Pixel:
-    Debug + Copy + Clone + Default + Send + Sync + PrimInt + 'static + private::Sealed
+    Sized
+    + Debug
+    + Copy
+    + Clone
+    + Default
+    + Send
+    + Sync
+    + Into<u16>
+    + From<u8>
+    + TryFrom<u16, Error: std::error::Error>
+    + 'static
+    + private::Sealed
 {
+    /// Conversion from `u16` to a pixel.
+    ///
+    /// If the pixel is a u8 (single byte), the conversion will be lossy.
+    /// Use [`From::<u8>::from`][From::from] instead if you want to create a pixel from an u8.
+    fn from_u16(value: u16) -> Self;
 }
 
 /// Pixel implementation for 8-bit video data.
 // SAFETY: u8 is valid if represented by a zeroed byte.
-unsafe impl Pixel for u8 {}
+unsafe impl Pixel for u8 {
+    fn from_u16(value: u16) -> Self {
+        value as Self
+    }
+}
 
 /// Pixel implementation for high bit-depth (9-16 bit) video data.
 // SAFETY: u16 is valid if represented by zeroed bytes.
-unsafe impl Pixel for u16 {}
+unsafe impl Pixel for u16 {
+    fn from_u16(value: u16) -> Self {
+        value
+    }
+}
diff --git a/src/pixel_ext.rs b/src/pixel_ext.rs
new file mode 100644
index 0000000..386144e
--- /dev/null
+++ b/src/pixel_ext.rs
@@ -0,0 +1,35 @@
+//! Extensions for the [`Pixel`] trait.
+//! 
+//! Currently only the [`PixelExt`] trait is provided, but more might be
+//! added in the future.
+
+use crate::pixel::Pixel;
+
+/// Implements additional traits on `Pixel` types for performance-sensitive applications.
+pub trait PixelExt {
+    /// Assume the reference is a reference to a `u8`.
+    /// 
+    /// # Safety
+    /// 
+    /// Callers must ensure that the implementing type is a `u8`, for example by
+    /// checking the size of the type (via `size_of::<T>() == 1`).
+    unsafe fn assume_u8(&self) -> &u8;
+
+    /// Assume the reference is a reference to a `u16`.
+    /// 
+    /// # Safety
+    /// 
+    /// Callers must ensure that the implementing type is a `u16`, for example by
+    /// checking the size of the type (via `size_of::<T>() == 2`).
+    unsafe fn assume_u16(&self) -> &u16;
+}
+
+impl<T: Pixel> PixelExt for T {
+    unsafe fn assume_u8(&self) -> &u8 {
+        unsafe { std::ptr::from_ref(self).cast::<u8>().as_ref_unchecked() }
+    }
+
+    unsafe fn assume_u16(&self) -> &u16 {
+        unsafe { std::ptr::from_ref(self).cast::<u16>().as_ref_unchecked() }
+    }
+}
diff --git a/src/plane.rs b/src/plane.rs
index bf701ee..b5eeb23 100644
--- a/src/plane.rs
+++ b/src/plane.rs
@@ -317,18 +317,8 @@ impl<T: Pixel> Plane<T> {
         let total = self.width().get() * self.height().get() * byte_width;
         ExactSizeWrapper {
             iter: self.pixels().flat_map(move |pix| {
-                let bytes: [u8; 2] = if byte_width == 1 {
-                    [
-                        pix.to_u8()
-                            .expect("Pixel::byte_data only supports u8 and u16 pixels"),
-                        0,
-                    ]
-                } else {
-                    pix.to_u16()
-                        .expect("Pixel::byte_data only supports u8 and u16 pixels")
-                        .to_le_bytes()
-                };
-                bytes.into_iter().take(byte_width)
+                let pix: u16 = pix.into();
+                pix.to_le_bytes().into_iter().take(byte_width)
             }),
             len: total,
         }