Derived clips: link MP4 versions, list them apart, video Enhance output

Groundwork for Enhance "Animate" (#208): a video made from a photo.

- Version linking reads the XMP link from MP4 / MOV too (top-level uuid box
  with the XMP UUID, mdat seeked over), at ingest and on Enhance approve.
- New kind `animated`, a derived clip (CLIP_KINDS): hidden behind the photo
  like a version, but not counted in photo_version_counts / version_count,
  never a display choice (400 clip_not_displayable), and listed as `clips`
  in GET /image/versions. Approve returns the clip entry.
- Templates whose imageapi:save is ImageApiSaveVideoVersion report
  output "video" in /enhance/templates (no enhance.toml change). Previews
  already relay ComfyUI's content type (video/mp4).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

CLAUDE.md

@@ -360,7 +360,12 @@ hash — design it alongside Branch D.
 
 A **version** is a non-destructive restoration, upscale or edit of a photo that
 an external process (the ComfyUI node in `comfyui-imageapi-versions`) writes next
-to the original, as `<dir>/aiedits/<original name>.<kind>[.<n>].png`. ImageApi
+to the original, as `<dir>/aiedits/<original name>.<kind>[.<n>].png`. A
+**derived clip** (kind `animated`, `CLIP_KINDS`) is a video made from a photo
+(Enhance "Animate", `<name>.animated[.<n>].mp4`): linked, hidden and lifecycled
+exactly like a version, but never counted as one (`photo_version_counts`,
+`version_count`), never a display choice (`PUT /image/versions/display` answers 400
+`clip_not_displayable`), and listed apart, as `clips` in `GET /image/versions`. ImageApi
 only detects it (library roots stay read-only) and shows original + versions as
 **one photo**; a *global* choice (not per user) says which file is displayed.
 Code: `src/database/version_dao.rs` (tables, DAO, SQL predicates),
@@ -376,8 +381,9 @@ for the `link_versions` binary).
   global choice; no row = the original, and a display file with no live row reads as
   the original.
 - **Linking at ingest** (`watcher::process_new_files`, before any derived work):
-  (1) a PNG's XMP `xmpMM:DerivedFrom/stRef:documentID = blake3:<hex>` (iTXt chunk
+  (1) the file's XMP `xmpMM:DerivedFrom/stRef:documentID = blake3:<hex>`: a PNG's
-  `XML:com.adobe.xmp`, read by hand, no decode); (2) the file-name convention next
+  iTXt chunk `XML:com.adobe.xmp`, or an MP4 / MOV's top-level `uuid` box with the XMP
+  UUID (both read by hand, no decode; `mdat` is seeked over); (2) the file-name convention next
   to an indexed original in the same library (same dir, or the parent of
   `aiedits/`). Name links are retried every tick (`sweep_name_links`, paged by a
   per-library resume cursor that wraps, so candidates that never resolve can't starve
@@ -455,7 +461,9 @@ auth and stays LAN-only, so ImageApi relays previews too.
   `file`, `prompt_mode`, `prompt_join`, `output_megapixels`).
 - **Templates** are API-format workflow exports, read-only (never edit them from
   here). Nodes are found by `_meta.title`: `imageapi:source` (required, gets the
-  photo path), `imageapi:save` (required, forced to `require_approval = true`),
+  photo path), `imageapi:save` (required, forced to `require_approval = true`; a
+  `ImageApiSaveVideoVersion` there makes the template's `output` `video`, an MP4 clip,
+  else `image`),
   `imageapi:prompt` (optional; fills the first of `text` / `prompt` / `value`),
   `imageapi:reference` (optional). `prompt_mode = "append"`: the user's text is joined
   after the template's own prompt with `prompt_join` (default one space); empty text
@@ -503,7 +511,7 @@ GET  /enhance/templates
       "templates": [{"name","label","description","accepts_prompt",
                      "prompt_mode": "append"|"replace", "prompt_required",
                      "accepts_reference","default","output_megapixels",
-                     "warn_above_megapixels"}]}
+                     "warn_above_megapixels", "output": "image"|"video"}]}
 POST /enhance  {"path","template","prompt"?,"reference_path"?,"library"?}  -> job
 GET  /enhance/jobs/{id}              -> job
 GET  /enhance/jobs?path=&library=    -> {"jobs": [job, ...]}  newest first, discarded hidden
@@ -511,8 +519,9 @@ GET  /enhance/prompts?templates=a,b&limit=10  (templates optional = all; limit 1
   -> {"prompts": [{"prompt","template","last_used","uses"}]}  distinct trimmed prompts, newest
      use first; template/last_used from the newest job; discarded + failed jobs count;
      DB only (no library scope, answers even when ComfyUI is down or Enhance is off)
-GET  /enhance/jobs/{id}/preview/{n}  -> the staged PNG (relayed from ComfyUI /view)
+GET  /enhance/jobs/{id}/preview/{n}  -> the staged PNG or MP4 (relayed from ComfyUI /view,
-POST /enhance/jobs/{id}/approve {"n"} -> a /image/versions `versions[]` entry
+                                        with its content type)
+POST /enhance/jobs/{id}/approve {"n"} -> a /image/versions `versions[]` (or `clips[]`) entry
 POST /enhance/jobs/{id}/discard      -> job (hidden from the list; ComfyUI untouched)
 
 job = {"id","status","queue_position"?,"error"?,"path","library_id","template",
@@ -737,7 +746,8 @@ POST   /image/tags/bulk  (many files, one transaction: { file_names, add: [names
 GET /image/versions?path=...&library=...   (path of the original or any version ->
                                             { original:{path,library_id,hash}, versions:[{path,
                                             library_id,hash,kind,source,created_at,references,
-                                            width,height}], display_hash })
+                                            width,height}], clips:[same shape, kind animated],
+                                            display_hash })
 POST /enhance ...                          (Enhance on demand; see "Enhance on demand")
 PUT /image/versions/display                (body { path, display_hash|null, library? }; global choice)
 //   /photos hides versions behind their original (include_versions=true lists them) and adds

src/database/version_dao.rs

@@ -20,7 +20,23 @@ use crate::database::{DbError, DbErrorKind, connect};
 use crate::otel::trace_db_call;
 
 /// The `kind` values a version may carry.
-pub const KINDS: &[&str] = &["restored", "upscale", "edit"];
+pub const KINDS: &[&str] = &["restored", "upscale", "edit", "animated"];
+
+/// Kinds that are *derived clips* rather than versions: a video made from a
+/// photo (Enhance "Animate"). Linked and hidden behind the photo like a
+/// version, but never counted as one, never a display choice, and listed
+/// separately (`clips` in `GET /image/versions`).
+pub const CLIP_KINDS: &[&str] = &["animated"];
+
+pub fn is_clip_kind(kind: &str) -> bool {
+    CLIP_KINDS.contains(&kind)
+}
+
+/// `pv.kind NOT IN (<clip kinds>)`, for queries about versions proper.
+fn not_a_clip_sql(alias: &str) -> String {
+    let kinds: Vec<String> = CLIP_KINDS.iter().map(|k| format!("'{k}'")).collect();
+    format!("{alias}.kind NOT IN ({})", kinds.join(", "))
+}
 
 /// `date_taken_source` on a version row whose date, GPS and camera fields
 /// were copied from its original.
@@ -731,16 +747,18 @@ impl VersionDao for SqliteVersionDao {
             }
             let mut conn = self.connection.lock().expect("Unable to get VersionDao");
 
-            let originals = sql_query(
+            let originals = sql_query(format!(
                 "SELECT o.content_hash AS original_hash, o.library_id AS library_id, \
                         o.rel_path AS rel_path, \
                         (SELECT COUNT(*) FROM photo_versions pv \
                           WHERE pv.original_hash = o.content_hash \
+                            AND {not_clip} \
                             AND EXISTS (SELECT 1 FROM image_exif v \
                                          WHERE v.content_hash = pv.version_hash)) AS version_count \
                  FROM image_exif o \
                  WHERE o.content_hash IN (SELECT original_hash FROM photo_versions)",
-            )
+                not_clip = not_a_clip_sql("pv"),
+            ))
             .load::<OriginalRow>(&mut *conn)
             .map_err(|e| anyhow::anyhow!("Query error: {}", e))?;
             if originals.is_empty() {
@@ -749,15 +767,17 @@ impl VersionDao for SqliteVersionDao {
 
             // Only a display choice that is still a version of that original
             // counts; anything else falls back to the original.
-            let displays = sql_query(
+            let displays = sql_query(format!(
                 "SELECT d.original_hash AS original_hash, v.library_id AS library_id, \
                         v.rel_path AS rel_path \
                  FROM photo_version_display d \
                  JOIN photo_versions pv ON pv.version_hash = d.display_hash \
                                        AND pv.original_hash = d.original_hash \
+                                       AND {not_clip} \
                  JOIN image_exif v ON v.content_hash = d.display_hash \
                  ORDER BY v.id",
-            )
+                not_clip = not_a_clip_sql("pv"),
+            ))
             .load::<DisplayRow>(&mut *conn)
             .map_err(|e| anyhow::anyhow!("Query error: {}", e))?;
             let mut display_by_hash: HashMap<String, Vec<(i32, String)>> = HashMap::new();

src/enhance/handlers.rs

@@ -221,6 +221,8 @@ struct TemplateView {
     /// Approximate output size; clients warn when the original is larger.
     output_megapixels: f64,
     warn_above_megapixels: f64,
+    /// `image` (a PNG version) or `video` (an MP4 clip, e.g. Animate).
+    output: super::template::OutputKind,
 }
 
 #[get("/enhance/templates")]
@@ -248,6 +250,7 @@ pub async fn get_templates(_: Claims, svc: web::Data<EnhanceService>) -> impl Re
             default: default.as_deref() == Some(t.cfg.name.as_str()),
             output_megapixels: t.cfg.output_megapixels,
             warn_above_megapixels: svc.config.warn_above_megapixels,
+            output: t.output,
         })
         .collect();
     let mut body = json!({
@@ -538,7 +541,13 @@ pub async fn approve(
         };
         build_response(dao.as_mut(), &original, job.library_id)
             .ok()
-            .and_then(|r| r.versions.into_iter().find(|v| v.hash == hash))
+            // An approved clip (Animate) is listed under `clips`.
+            .and_then(|r| {
+                r.versions
+                    .into_iter()
+                    .chain(r.clips)
+                    .find(|v| v.hash == hash)
+            })
     };
     match entry {
         Some(v) => HttpResponse::Ok().json(v),

src/enhance/template.rs

@@ -19,6 +19,19 @@ pub const TITLE_REFERENCE: &str = "imageapi:reference";
 /// Prompt-node inputs, in the order they are tried.
 const PROMPT_INPUTS: &[&str] = &["text", "prompt", "value"];
 
+/// What a template produces, from its `imageapi:save` node's class.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize)]
+#[serde(rename_all = "lowercase")]
+pub enum OutputKind {
+    /// Save Image Version: PNG versions.
+    Image,
+    /// Save Video Version: an MP4 (e.g. a clip animated from the photo).
+    Video,
+}
+
+/// The save node class that makes a template's output a video.
+pub const VIDEO_SAVE_CLASS: &str = "ImageApiSaveVideoVersion";
+
 /// Output nodes that write into ComfyUI's output folder on every API run.
 const STRAY_OUTPUT_CLASSES: &[&str] = &[
     "SaveImage",
@@ -38,6 +51,8 @@ pub struct Template {
     /// `(node id, input name)` of the prompt text.
     prompt: Option<(String, &'static str)>,
     reference_id: Option<String>,
+    /// Image or video, from the save node's class.
+    pub output: OutputKind,
     /// Startup warnings (stray output nodes).
     pub warnings: Vec<String>,
 }
@@ -143,6 +158,11 @@ impl Template {
                 ));
             }
         }
+        let output = if workflow[&save_id]["class_type"].as_str() == Some(VIDEO_SAVE_CLASS) {
+            OutputKind::Video
+        } else {
+            OutputKind::Image
+        };
         Ok(Template {
             cfg,
             workflow,
@@ -150,6 +170,7 @@ impl Template {
             save_id,
             prompt,
             reference_id,
+            output,
             warnings,
         })
     }
@@ -414,6 +435,23 @@ mod tests {
         assert_ne!(input(&x.prompt, "5", "seed"), input(&y.prompt, "5", "seed"));
     }
 
+    #[test]
+    fn output_kind_follows_the_save_node() {
+        assert_eq!(
+            load(RESTORE, PromptMode::Append).unwrap().output,
+            OutputKind::Image
+        );
+        let video = RESTORE.replace(
+            "\"ImageApiSaveImageVersion\"",
+            "\"ImageApiSaveVideoVersion\"",
+        );
+        assert_ne!(video, RESTORE);
+        assert_eq!(
+            load(&video, PromptMode::Append).unwrap().output,
+            OutputKind::Video
+        );
+    }
+
     #[test]
     fn random_seeds_fit_seedvr2s_u32_limit() {
         // SeedVR2VideoUpscaler rejects seeds above 2^32 - 1 at validation.

src/versions/handlers.rs

@@ -12,7 +12,7 @@ use serde::{Deserialize, Serialize};
 
 use crate::data::Claims;
 use crate::database::ExifDao;
-use crate::database::version_dao::VersionDao;
+use crate::database::version_dao::{VersionDao, is_clip_kind};
 use crate::libraries;
 use crate::state::AppState;
 use crate::utils::normalize_path;
@@ -66,6 +66,9 @@ pub struct VersionsResponse {
     /// standalone photos).
     pub original: Option<FileRef>,
     pub versions: Vec<VersionEntry>,
+    /// Derived clips: videos made from the photo (kind `animated`). Same
+    /// shape as `versions`, but never a display choice.
+    pub clips: Vec<VersionEntry>,
     /// The globally displayed version; `null` = the original.
     pub display_hash: Option<String>,
 }
@@ -137,6 +140,7 @@ pub fn build_response(
 
     let files = dao.list_versions(original_hash)?;
     let mut versions = Vec::with_capacity(files.len());
+    let mut clips = Vec::new();
     for f in files {
         let reference_hashes = dao.list_references(&f.version_hash)?;
         let located = dao.files_for_hashes(&reference_hashes, Some(f.library_id))?;
@@ -151,7 +155,12 @@ pub fn build_response(
                 }
             })
             .collect();
-        versions.push(VersionEntry {
+        let list = if is_clip_kind(&f.kind) {
+            &mut clips
+        } else {
+            &mut versions
+        };
+        list.push(VersionEntry {
             path: f.rel_path,
             library_id: f.library_id,
             hash: f.version_hash,
@@ -177,6 +186,7 @@ pub fn build_response(
     Ok(VersionsResponse {
         original,
         versions,
+        clips,
         display_hash,
     })
 }
@@ -241,6 +251,13 @@ pub async fn put_display(
                     return ApiError::internal("internal error").into_response();
                 }
             };
+        if current.clips.iter().any(|c| c.hash == hash) {
+            return ApiError::bad_request(
+                "clip_not_displayable",
+                "A clip made from this photo can't be shown in its place",
+            )
+            .into_response();
+        }
         if !current.versions.iter().any(|v| v.hash == hash) {
             return ApiError::bad_request(
                 "not_a_version",

src/versions/integration_tests.rs

@@ -14,7 +14,7 @@ use diesel::sql_query;
 use diesel::sql_types::{BigInt, Text};
 use diesel::sqlite::SqliteConnection;
 
-use super::xmp::test_png::{producer_xmp, write_png};
+use super::xmp::test_png::{mp4_bytes, producer_xmp, write_png};
 use super::*;
 use crate::data::Claims;
 use crate::database::test::in_memory_db_connection;
@@ -782,6 +782,49 @@ async fn display_choice_set_from_a_version_path_cleared_and_validated() {
     assert_eq!(awtest::call_service(&app, req).await.status(), 200);
 }
 
+#[actix_rt::test]
+async fn an_animated_clip_links_hides_and_is_listed_apart_from_versions() {
+    let fx = fx();
+    let original = fx.put("2014/IMG_1.JPG", b"photo bytes");
+    let rel = "2014/aiedits/IMG_1.JPG.animated.mp4";
+    let clip = fx.put(
+        rel,
+        &mp4_bytes(Some(&producer_xmp(&original, "animated", &[])), 3),
+    );
+    let files = vec![(fx.root.join(rel), rel.to_string())];
+    assert_eq!(
+        link_new_files(&fx.library, &files, &fx.exif_arc(), &fx.state.version_dao),
+        1
+    );
+    let app = photos_app!(fx);
+
+    // Hidden behind the photo, like a version, and not counted as one.
+    let (_, body) = get_json!(app, "/photos?path=&recursive=true&sort=nameasc");
+    assert_eq!(body["photos"], serde_json::json!(["2014/IMG_1.JPG"]));
+    assert!(body.get("photo_version_counts").is_none(), "{body}");
+
+    // Listed under `clips`, from the photo's path or the clip's own.
+    for path in ["2014/IMG_1.JPG", rel] {
+        let (status, body) = get_json!(app, &format!("/image/versions?path={path}"));
+        assert_eq!(status, 200);
+        assert_eq!(body["versions"], serde_json::json!([]));
+        assert_eq!(body["clips"][0]["hash"], clip);
+        assert_eq!(body["clips"][0]["kind"], "animated");
+        assert_eq!(body["clips"][0]["path"], rel);
+    }
+
+    // Never the photo's display version.
+    let req = awtest::TestRequest::put()
+        .uri("/image/versions/display")
+        .insert_header(("Authorization", format!("Bearer {}", Fx::token())))
+        .set_json(serde_json::json!({"path": "2014/IMG_1.JPG", "display_hash": clip}))
+        .to_request();
+    let resp = awtest::call_service(&app, req).await;
+    assert_eq!(resp.status(), 400);
+    let body: serde_json::Value = awtest::read_body_json(resp).await;
+    assert_eq!(body["code"], "clip_not_displayable");
+}
+
 #[actix_rt::test]
 async fn deleting_the_displayed_version_falls_back_to_the_original() {
     let (fx, _o, v1, _v2) = seeded();

src/versions/mod.rs

@@ -6,7 +6,7 @@
 //! read-only rule for library roots holds. This module owns
 //!
 //! - the link rules at ingest ([`link_new_files`], [`sweep_name_links`]):
-//!   the PNG's XMP `DerivedFrom` hash first, then the
+//!   the XMP `DerivedFrom` hash first (PNG, or MP4 / MOV for a clip), then the
 //!   `<name>.<kind>[.<n>].<ext>` file-name convention;
 //! - the HTTP surface (`GET /image/versions`, `PUT /image/versions/display`).
 //!
@@ -132,18 +132,16 @@ pub fn name_link(
     })
 }
 
-/// Resolve the XMP link of a PNG on disk. `None` when the file isn't a PNG,
+/// Resolve the XMP link of a version file on disk (PNG, or MP4 / MOV for a
-/// can't be read, or carries no `DerivedFrom` blake3 id.
+/// derived clip). `None` when the file isn't one of those, can't be read, or
+/// carries no `DerivedFrom` blake3 id.
 pub fn xmp_link(
     library: &Library,
     rel_path: &str,
     abs_path: &Path,
     version_hash: &str,
 ) -> Option<NewVersionLink> {
-    let is_png = abs_path
+    if !xmp::has_readable_xmp(abs_path) {
-        .extension()
-        .is_some_and(|e| e.eq_ignore_ascii_case("png"));
-    if !is_png {
         return None;
     }
     let parsed = match xmp::read_version_xmp(abs_path) {
@@ -213,14 +211,14 @@ pub fn link_new_files(
     exif_dao: &Arc<Mutex<Box<dyn ExifDao>>>,
     version_dao: &Arc<Mutex<Box<dyn VersionDao>>>,
 ) -> usize {
-    let pngs: Vec<&(PathBuf, String)> = files
+    let candidates: Vec<&(PathBuf, String)> = files
         .iter()
-        .filter(|(p, _)| p.extension().is_some_and(|e| e.eq_ignore_ascii_case("png")))
+        .filter(|(p, _)| xmp::has_readable_xmp(p))
         .collect();
-    if pngs.is_empty() {
+    if candidates.is_empty() {
         return 0;
     }
-    let rels: Vec<String> = pngs.iter().map(|(_, r)| r.clone()).collect();
+    let rels: Vec<String> = candidates.iter().map(|(_, r)| r.clone()).collect();
     // Chunked: a first scan can register far more paths than SQLite's
     // bound-variable limit allows in one `IN (...)`.
     let mut rows = Vec::with_capacity(rels.len());
@@ -245,7 +243,7 @@ pub fn link_new_files(
         .collect();
 
     let mut linked = 0;
-    for (abs, rel) in pngs {
+    for (abs, rel) in candidates {
         let Some(hash) = hashes.get(rel.as_str()) else {
             continue;
         };

src/versions/xmp.rs

@@ -1,6 +1,8 @@
 //! Reading the version link out of a PNG: the `XML:com.adobe.xmp` iTXt
 //! chunk the ComfyUI producer writes (`comfyui-imageapi-versions`), and the
 //! ComfyUI `prompt` tEXt chunk the backfill uses for files without XMP.
+//! Video versions (MP4 / MOV, e.g. a clip animated from a photo) carry the
+//! same packet in a top-level `uuid` box with the XMP UUID.
 //!
 //! The chunk stream is walked by hand: no pixel decode, and the `png`
 //! crate's text API would still inflate every IDAT's neighbours for no gain.
@@ -161,9 +163,77 @@ pub fn parse_version_xmp(xmp: &str) -> Option<VersionXmp> {
     })
 }
 
-/// Parse the version link out of a PNG file; `None` if it has none.
+/// The XMP UUID (Adobe XMP spec part 3): a top-level `uuid` box with this
+/// id holds the packet in MP4 / MOV files.
+const XMP_UUID: [u8; 16] = [
+    0xbe, 0x7a, 0xcf, 0xcb, 0x97, 0xa9, 0x42, 0xe8, 0x9c, 0x71, 0x99, 0x94, 0x91, 0xe3, 0xaf, 0xac,
+];
+
+/// The XMP packet of an MP4 / MOV file, if it has one. Walks the top-level
+/// box headers only, seeking over `mdat`, so a large clip costs a few small
+/// reads. `Ok(None)` for a file that isn't box-structured.
+pub fn read_mp4_xmp(path: &Path) -> io::Result<Option<String>> {
+    let mut reader = BufReader::new(File::open(path)?);
+    let file_len = reader.get_ref().metadata()?.len();
+    let mut pos: u64 = 0;
+    while pos + 8 <= file_len {
+        let mut head = [0u8; 8];
+        reader.read_exact(&mut head)?;
+        let mut size = u64::from(u32::from_be_bytes([head[0], head[1], head[2], head[3]]));
+        let kind = [head[4], head[5], head[6], head[7]];
+        let mut header_len = 8u64;
+        if size == 1 {
+            let mut big = [0u8; 8];
+            reader.read_exact(&mut big)?;
+            size = u64::from_be_bytes(big);
+            header_len = 16;
+        } else if size == 0 {
+            size = file_len - pos; // runs to the end of the file
+        }
+        if size < header_len || pos + size > file_len {
+            return Ok(None); // not an MP4, or truncated: no link
+        }
+        let body = size - header_len;
+        if &kind == b"uuid" && body >= 16 && body - 16 <= u64::from(MAX_TEXT_CHUNK) {
+            let mut id = [0u8; 16];
+            reader.read_exact(&mut id)?;
+            if id == XMP_UUID {
+                let mut data = vec![0u8; (body - 16) as usize];
+                reader.read_exact(&mut data)?;
+                return Ok(Some(String::from_utf8_lossy(&data).into_owned()));
+            }
+            reader.seek_relative((body - 16) as i64)?;
+        } else {
+            reader.seek(SeekFrom::Current(body as i64))?;
+        }
+        pos += size;
+    }
+    Ok(None)
+}
+
+/// File types a version's XMP link is read from.
+pub fn has_readable_xmp(path: &Path) -> bool {
+    path.extension().and_then(|e| e.to_str()).is_some_and(|e| {
+        ["png", "mp4", "mov"]
+            .iter()
+            .any(|x| e.eq_ignore_ascii_case(x))
+    })
+}
+
+/// Parse the version link out of a PNG, MP4 or MOV file; `None` if it has
+/// none (or isn't one of those).
 pub fn read_version_xmp(path: &Path) -> io::Result<Option<VersionXmp>> {
-    Ok(read_png_xmp(path)?.and_then(|x| parse_version_xmp(&x)))
+    let is_png = path
+        .extension()
+        .is_some_and(|e| e.eq_ignore_ascii_case("png"));
+    let xmp = if is_png {
+        read_png_xmp(path)?
+    } else if has_readable_xmp(path) {
+        read_mp4_xmp(path)?
+    } else {
+        None
+    };
+    Ok(xmp.and_then(|x| parse_version_xmp(&x)))
 }
 
 #[cfg(test)]
@@ -177,6 +247,28 @@ pub(crate) mod test_png {
         out.extend_from_slice(&[0, 0, 0, 0]); // CRC: the reader never checks it
     }
 
+    fn mp4_box(out: &mut Vec<u8>, kind: &[u8; 4], data: &[u8]) {
+        out.extend_from_slice(&((data.len() + 8) as u32).to_be_bytes());
+        out.extend_from_slice(kind);
+        out.extend_from_slice(data);
+    }
+
+    /// An MP4-shaped byte stream as the producer writes it: ftyp, moov,
+    /// mdat, then the optional XMP `uuid` box. Not playable; enough for the
+    /// box walker (unique per `salt`).
+    pub fn mp4_bytes(xmp: Option<&str>, salt: u8) -> Vec<u8> {
+        let mut out = Vec::new();
+        mp4_box(&mut out, b"ftyp", b"isom\0\0\x02\0");
+        mp4_box(&mut out, b"moov", &[1; 24]);
+        mp4_box(&mut out, b"mdat", &[salt; 64]);
+        if let Some(x) = xmp {
+            let mut d = super::XMP_UUID.to_vec();
+            d.extend_from_slice(x.as_bytes());
+            mp4_box(&mut out, b"uuid", &d);
+        }
+        out
+    }
+
     /// A PNG-shaped byte stream: signature, IHDR, optional tEXt `prompt`,
     /// optional XMP iTXt, an IDAT, IEND. Not decodable; enough for the
     /// chunk walker (and unique per `salt`, so each file hashes differently).
@@ -247,6 +339,56 @@ mod tests {
     /// The packet embedded in a real archive restoration (copied verbatim).
     const REAL_XMP: &str = r#"<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?><x:xmpmeta xmlns:x="adobe:ns:meta/"><rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"><rdf:Description rdf:about="" xmlns:xmpMM="http://ns.adobe.com/xap/1.0/mm/" xmlns:stRef="http://ns.adobe.com/xap/1.0/sType/ResourceRef#" xmlns:imageapi="https://git.crkc.xyz/Apps/ImageApi/ns/versions/1.0/"><xmpMM:DerivedFrom rdf:parseType="Resource"><stRef:documentID>blake3:d5fa14088629f680f4f8e5815df473cc4684a97fc937da588baf72d2693dea3a</stRef:documentID><stRef:filePath>019_19.JPG</stRef:filePath></xmpMM:DerivedFrom><imageapi:versionKind>restored</imageapi:versionKind></rdf:Description></rdf:RDF></x:xmpmeta><?xpacket end="w"?>"#;
 
+    #[test]
+    fn reads_the_link_from_an_mp4_clip() {
+        let dir = tempfile::tempdir().unwrap();
+        let clip = dir.path().join("IMG_1.JPG.animated.mp4");
+        let xmp = REAL_XMP.replace("restored", "animated");
+        std::fs::write(&clip, mp4_bytes(Some(&xmp), 1)).unwrap();
+        let v = read_version_xmp(&clip).unwrap().unwrap();
+        assert_eq!(v.original_hash, ORIG);
+        assert_eq!(v.kind.as_deref(), Some("animated"));
+        // Same bytes under .mov are read too; other extensions aren't.
+        let mov = dir.path().join("clip.MOV");
+        std::fs::write(&mov, mp4_bytes(Some(&xmp), 1)).unwrap();
+        assert!(read_version_xmp(&mov).unwrap().is_some());
+        let avi = dir.path().join("clip.avi");
+        std::fs::write(&avi, mp4_bytes(Some(&xmp), 1)).unwrap();
+        assert!(read_version_xmp(&avi).unwrap().is_none());
+    }
+
+    #[test]
+    fn mp4_without_xmp_or_not_an_mp4_has_no_link() {
+        let dir = tempfile::tempdir().unwrap();
+        let plain = dir.path().join("plain.mp4");
+        std::fs::write(&plain, mp4_bytes(None, 2)).unwrap();
+        assert_eq!(read_mp4_xmp(&plain).unwrap(), None);
+        let junk = dir.path().join("junk.mp4");
+        std::fs::write(&junk, b"definitely not boxes at all").unwrap();
+        assert_eq!(read_mp4_xmp(&junk).unwrap(), None);
+    }
+
+    #[test]
+    fn mp4_64bit_box_sizes_are_walked() {
+        let dir = tempfile::tempdir().unwrap();
+        let mut data = Vec::new();
+        data.extend_from_slice(&16u32.to_be_bytes());
+        data.extend_from_slice(b"ftypisom\0\0\x02\0");
+        // mdat with a 64-bit size field
+        data.extend_from_slice(&1u32.to_be_bytes());
+        data.extend_from_slice(b"mdat");
+        data.extend_from_slice(&(16u64 + 8).to_be_bytes());
+        data.extend_from_slice(&[7; 8]);
+        let mut d = XMP_UUID.to_vec();
+        d.extend_from_slice(REAL_XMP.as_bytes());
+        data.extend_from_slice(&((d.len() + 8) as u32).to_be_bytes());
+        data.extend_from_slice(b"uuid");
+        data.extend_from_slice(&d);
+        let f = dir.path().join("big.mp4");
+        std::fs::write(&f, data).unwrap();
+        assert_eq!(read_mp4_xmp(&f).unwrap().as_deref(), Some(REAL_XMP));
+    }
+
     #[test]
     fn parses_the_real_producer_packet() {
         let x = parse_version_xmp(REAL_XMP).unwrap();