diff --git a/crates/engine/src/media.rs b/crates/engine/src/media.rs
index 6f6f0359..6c917b23 100644
--- a/crates/engine/src/media.rs
+++ b/crates/engine/src/media.rs
@@ -803,7 +803,31 @@ impl RenderJob {
         self
     }
 
+    /// Render. A panic inside (a decoder or pipeline bug) is this job's failure: the view waiting
+    /// for it is told, instead of waiting forever on a worker that died (the pool doesn't replace
+    /// it, so a few such panics used to stop every render).
     pub fn run(self) -> RenderResult {
+        let failed = self.failed(String::new());
+        crate::guard::catch("the render", || self.run_unguarded()).unwrap_or_else(|e| RenderResult { rendered: Err(e), ..failed })
+    }
+
+    /// This job's result when it fails with `e`.
+    fn failed(&self, e: String) -> RenderResult {
+        RenderResult {
+            request_id: self.request_id,
+            source_key: self.source_key,
+            photo: self.photo,
+            level: self.level,
+            key: self.key,
+            rendered: Err(e),
+            gpu_fallback: None,
+            loaded: None,
+            quick: None,
+            display: None,
+        }
+    }
+
+    fn run_unguarded(self) -> RenderResult {
         if let Some((cache, key)) = &self.cache
             && let Some(img) = cache.get_at(self.cache_generation, *key)
         {
@@ -877,18 +901,7 @@ impl RenderJob {
                     display,
                 }
             }
-            Err(e) => RenderResult {
-                request_id: self.request_id,
-                source_key: self.source_key,
-                photo: self.photo,
-                level: self.level,
-                key: self.key,
-                rendered: Err(e),
-                gpu_fallback: None,
-                loaded: None,
-                quick: None,
-                display: None,
-            },
+            Err(e) => self.failed(e),
         }
     }
 }
@@ -976,7 +989,25 @@ pub struct QuickJob {
 }
 
 impl QuickJob {
+    /// The stand-in, from the first source that has one. A panic inside is this job's failure
+    /// (see [`RenderJob::run`]).
     pub fn run(self) -> RenderResult {
+        let (request_id, photo, key) = (self.request_id, self.photo, self.key);
+        crate::guard::catch("the render", || self.run_unguarded()).unwrap_or_else(|e| RenderResult {
+            request_id,
+            source_key: None,
+            photo,
+            level: SourceLevel::Thumb,
+            key,
+            rendered: Err(e),
+            gpu_fallback: None,
+            loaded: None,
+            quick: None,
+            display: None,
+        })
+    }
+
+    fn run_unguarded(self) -> RenderResult {
         let (photo, key) = (self.photo, self.key);
         let request_id = self.request_id;
         let done = |rendered: Result<Rendered, String>, quick| RenderResult {
@@ -1588,6 +1619,31 @@ mod thumbnail_hash_tests {
         s.accept(&old);
         assert_eq!(s.media.source_usage().0, 0);
     }
+    /// A panic while rendering (here in the decoder) used to end the pool's worker thread without
+    /// an answer, so the view that asked showed "Rendering…" forever. It is the job's failure now,
+    /// for renders and for stand-ins, answering the request it came from.
+    #[test]
+    fn a_render_that_panics_fails_instead_of_never_answering() {
+        let mut s = session();
+        s.media.file_loader = Some(Arc::new(|_, _| panic!("decoder bug")));
+        let mut job = s.thumb_job(PhotoId(1), 256).unwrap();
+        job.request_id = 42;
+        let (key, level, source_key) = (job.key, job.level, job.source_key);
+        let r = job.run();
+        assert_eq!((r.request_id, r.photo, r.key, r.level, r.source_key), (42, PhotoId(1), key, level, source_key));
+        assert_eq!(r.rendered.err().as_deref(), Some("the render failed unexpectedly: decoder bug"));
+        let quick = QuickJob {
+            request_id: 7,
+            photo: PhotoId(1),
+            key: 9,
+            cached: Vec::new(),
+            embedded: Some(("x.arw".into(), Arc::new(|_: &str, _: usize| -> Option<Rgba8> { panic!("preview reader bug") }), 256)),
+            small: None,
+        };
+        let r = quick.run();
+        assert_eq!((r.request_id, r.key), (7, 9));
+        assert_eq!(r.rendered.err().as_deref(), Some("the render failed unexpectedly: preview reader bug"));
+    }
 }
 
 #[cfg(test)]
diff --git a/crates/ui-egui/src/panels/dialogs.rs b/crates/ui-egui/src/panels/dialogs.rs
index 72528fca..d94bc619 100644
--- a/crates/ui-egui/src/panels/dialogs.rs
+++ b/crates/ui-egui/src/panels/dialogs.rs
@@ -98,8 +98,8 @@ fn watermark_preview(
                 let (out, repaint) = (base.clone(), ctx.clone());
                 let work = move || {
                     for job in jobs {
-                        let r = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| job.run().rendered.map(|r| r.image)))
-                            .unwrap_or_else(|_| Err("the preview failed unexpectedly".into()));
+                        // (a panic in the render is its failure: `RenderJob::run`)
+                        let r = job.run().rendered.map(|r| r.image);
                         let mut got = out.lock().unwrap_or_else(std::sync::PoisonError::into_inner);
                         // a failed sharp render keeps the thumbnail
                         if r.is_ok() || !matches!(got.1, Some(Ok(_))) {
diff --git a/crates/ui-egui/src/render.rs b/crates/ui-egui/src/render.rs
index 66bce01f..03ac1894 100644
--- a/crates/ui-egui/src/render.rs
+++ b/crates/ui-egui/src/render.rs
@@ -1057,6 +1057,39 @@ mod thumbnail_tests {
         let _ = std::fs::remove_dir_all(&lib);
     }
 
+    /// A render that panics (a decoder bug) is a failure like any other: the slot stops waiting,
+    /// the error is kept (not retried every frame), and the pool's workers live on. Before, the
+    /// worker died without an answer and the slot stayed pending forever.
+    #[test]
+    fn a_panicking_render_is_a_failure_not_a_dead_worker() {
+        let (mut app, _, ctx) = fixture_with(Session::new(), (*photo()).clone());
+        app.session.media.file_loader = Some(Arc::new(|_, _| panic!("decoder bug")));
+        app.renderer.offload = None; // the real job pool
+        let slot = Slot::Thumb(PhotoId(1));
+        let ask = |app: &mut crate::LightcraftApp| {
+            if let Some(job) = app.session.thumb_job(PhotoId(1), 256) {
+                app.renderer.request(slot, job, 30);
+            }
+            app.renderer.poll(&ctx, &mut app.session);
+        };
+        ask(&mut app);
+        let t0 = std::time::Instant::now();
+        while app.renderer.is_pending(slot) {
+            assert!(t0.elapsed() < std::time::Duration::from_secs(20), "the render never answered");
+            std::thread::sleep(std::time::Duration::from_millis(5));
+            app.renderer.poll(&ctx, &mut app.session);
+        }
+        assert_eq!(app.renderer.failure(slot), Some("the render failed unexpectedly: decoder bug"));
+        let workers = app.renderer.live_workers();
+        assert!(workers > 0, "the worker that panicked is still there");
+        let done = app.renderer.completed;
+        for _ in 0..30 {
+            ask(&mut app);
+        }
+        assert!(!app.renderer.is_pending(slot) && app.renderer.completed == done, "no retry loop");
+        assert_eq!(app.renderer.live_workers(), workers);
+    }
+
     #[test]
     fn raw_quick_path_frames_build_no_jobs() {
         // an unedited raw: the grid shows its embedded preview (held here until released) and stops
diff --git a/docs/whats-new.md b/docs/whats-new.md
index fb8b285a..60c470d7 100644
--- a/docs/whats-new.md
+++ b/docs/whats-new.md
@@ -45,6 +45,10 @@
 - Upgrading from a Windows installer older than 0.6 (such as 0.5.0) keeps LightCraft in the folder it was installed
   in, instead of moving it back to Program Files (issue #399). If that folder has been deleted since, the upgrade
   installs to Program Files.
+- A photo that runs into a bug in LightCraft while it renders (in a raw decoder or the editing pipeline) now fails
+  like a photo that can't be opened, with the reason, instead of rendering forever; its thumbnail stops waiting too.
+  Each such bug used to take one of the render threads with it, so after a few of them no photo rendered until
+  LightCraft was restarted. The bug is still written to the log (Help ▸ Open Log Folder) for a report.
 
 ## 2026-10-10