diff --git a/crates/engine/src/cmd/create.rs b/crates/engine/src/cmd/create.rs
index 6ec7d4e59..3a9f6d67f 100644
--- a/crates/engine/src/cmd/create.rs
+++ b/crates/engine/src/cmd/create.rs
@@ -362,15 +362,13 @@ fn free_spot(s: &Session, w: f64, h: f64) -> Result<(f64, f64)> {
     })
 }
 
-/// The bytes of the file at `path`, within the session's allowed file roots.
 #[cfg(not(target_arch = "wasm32"))]
-pub(super) fn read_path(path: &str) -> Result<Vec<u8>> {
+fn read_path(path: &str) -> Result<Vec<u8>> {
     crate::file_access::read(path)
 }
 
-/// The browser has no files to read: always an error.
 #[cfg(target_arch = "wasm32")]
-pub(super) fn read_path(path: &str) -> Result<Vec<u8>> {
+fn read_path(path: &str) -> Result<Vec<u8>> {
     Err(crate::EngineError::Other(format!("{path}: files can't be read in the browser")))
 }
 
diff --git a/crates/engine/src/cmd/text.rs b/crates/engine/src/cmd/text.rs
index 0f0f003de..6e1f5c3db 100644
--- a/crates/engine/src/cmd/text.rs
+++ b/crates/engine/src/cmd/text.rs
@@ -497,7 +497,7 @@ pub fn specs() -> Vec<CommandSpec> {
                     let b64 = b64.split_once(',').map(|(_, d)| d.to_string()).unwrap_or(b64);
                     base64::engine::general_purpose::STANDARD.decode(b64.trim()).map_err(|e| bad(cmd, format!("`data`: {e}")))?
                 } else if let Some(path) = string(p, "path") {
-                    super::create::read_path(&path)?
+                    super::file::read_capped(&path, figcraft_text::fontdb::MAX_FONT_FILE)?
                 } else {
                     return Err(bad(cmd, "pass `path` or `data` (base64)"));
                 };
diff --git a/crates/text/src/fontdb.rs b/crates/text/src/fontdb.rs
index 814d3a2f7..604f3eee5 100644
--- a/crates/text/src/fontdb.rs
+++ b/crates/text/src/fontdb.rs
@@ -819,13 +819,19 @@ fn scan_dir(dir: &std::path::Path, depth: usize, out: &mut BTreeMap<String, (Str
         // Very large files are indexed all the same (Apple Color Emoji is ~180 MB and is where the
         // system's colour emoji live): `index_bytes` reads headers and name tables out of a zeroed
         // buffer, so only those parts are ever touched.
-        let Some(bytes) = usize::try_from(meta.len()).ok().filter(|l| *l <= 256 * 1024 * 1024).and_then(|l| index_bytes(&path, l)) else { continue };
+        if meta.len() > MAX_FONT_FILE {
+            continue;
+        }
+        let Some(bytes) = usize::try_from(meta.len()).ok().and_then(|l| index_bytes(&path, l)) else { continue };
         for (index, family, style) in faces_in(&bytes) {
             out.entry(family.to_lowercase()).or_insert_with(|| (family.clone(), Vec::new())).1.push(Pending { path: path.clone(), index, style });
         }
     }
 }
 
+/// The largest font file read (CJK collections run to ~120 MB, Apple Color Emoji ~180 MB).
+pub const MAX_FONT_FILE: u64 = 256 * 1024 * 1024;
+
 /// The parts of a font file the scan reads: headers, table directories and `name` tables, at their
 /// offsets in a zeroed buffer of the file's length (reading whole files was most of the scan).
 fn index_bytes(path: &std::path::Path, len: usize) -> Option<Vec<u8>> {