Fix named reference lookup for long videos

dumas_h3_longvideos.py

@@ -4463,23 +4463,51 @@ def _slot_refs_for_text(text, ref_slots):
 
 
 def _named_character_refs_for_text(text, ref_slots):
+    return _named_refs_for_text(text, ref_slots, kinds=("character",))
+
+
+def _named_refs_for_text(text, ref_slots, kinds=("character",)):
+    """Backward-compatible alias for older call sites.
+
+    The script annotation path still expects the shorter helper name, while the
+    newer reference matcher keeps the more explicit character-specific helper.
+    """
+    kinds = tuple(dict.fromkeys(kinds or ()))
+    if not kinds:
+        return []
     haystack = str(text or "")
-    matched = []
+    kind_hits = {kind: [] for kind in kinds}
+    name_hits = {kind: {} for kind in kinds}
     for slot_number, raw in enumerate(ref_slots or [], 1):
         ref = _reference_slot(raw, slot_number)
-        if ref.get("kind") != "character" or _reference_image(ref) is None:
+        kind = ref.get("kind")
+        if kind not in kind_hits or _reference_image(ref) is None:
             continue
+        matched_names = []
         for name in _reference_name_keys(ref):
             if re.search(r"\b" + re.escape(name) + r"\b", haystack, re.I):
+                matched_names.append(name.lower())
+                name_hits[kind].setdefault(name.lower(), []).append((slot_number, ref))
+        if matched_names:
+            kind_hits[kind].append((slot_number, ref, tuple(set(matched_names))))
+    matched = []
+    seen = set()
+    for kind in kinds:
+        for slot_number, ref, matched_names in kind_hits.get(kind, []):
+            if not any(len(name_hits[kind].get(name, [])) == 1 for name in matched_names):
+                continue
+            marker = id(ref)
+            if marker in seen:
+                continue
+            seen.add(marker)
             matched.append((slot_number, ref))
-                break
     return matched
 
 
 def _matched_reference_slots(text, ref_slots):
     matched = []
     seen = set()
-    for slot_number, ref in _slot_refs_for_text(text, ref_slots) + _named_character_refs_for_text(text, ref_slots):
+    for slot_number, ref in _slot_refs_for_text(text, ref_slots) + _named_refs_for_text(text, ref_slots, kinds=("character", "location")):
         if slot_number in seen:
             continue
         seen.add(slot_number)
@@ -4487,7 +4515,7 @@ def _matched_reference_slots(text, ref_slots):
     return matched
 
 
-def _reference_context_for_text(text, ref_slots):
+def _reference_context_for_text(text, ref_slots, include_character_wardrobe=True):
     parts = []
     for slot_number, ref in _matched_reference_slots(text, ref_slots):
         label = _reference_text(ref.get("name")) or _reference_text(ref.get("id")) or f"reference {slot_number}"
@@ -4505,7 +4533,7 @@ def _reference_context_for_text(text, ref_slots):
             parts.append(facts)
         if description:
             parts.append(f"Persistent appearance for {label}: {description}")
-        if wardrobe:
+        if include_character_wardrobe and wardrobe:
             parts.append(f"Persistent wardrobe/style for {label}: {wardrobe}")
         if general:
             parts.append(f"Character notes for {label}: {general}")
@@ -4570,7 +4598,7 @@ def resolve_tagged_refs(text, ref_list):
     A tag naming a slot with no image connected refers to nothing at all, so it is
     removed from the text rather than left to confuse the encoder, and reported."""
     wanted = picture_tags(text)
-    live = [n for n in wanted if 1 <= n <= len(ref_list or []) and ref_list[n - 1] is not None]
+    live = [n for n in wanted if 1 <= n <= len(ref_list or []) and _reference_image(ref_list[n - 1]) is not None]
     dropped = [n for n in wanted if n not in live]
     renumber = {old: new for new, old in enumerate(live, 1)}
 
@@ -4586,7 +4614,7 @@ def resolve_tagged_refs(text, ref_list):
     return out.strip(), [ref_list[n - 1] for n in live], dropped
 
 
-def resolve_prompt_refs(text, ref_list):
+def resolve_prompt_refs(text, ref_list, include_named=True):
     """(rewritten text, refs, dropped) for the refs a shot actually carries.
 
     Explicit <Picture N> tags still decide which slot numbers the prompt points at,
@@ -4597,7 +4625,8 @@ def resolve_prompt_refs(text, ref_list):
     rewritten, tagged_refs, dropped = resolve_tagged_refs(text, ref_list)
     refs = list(tagged_refs)
     seen = {id(ref) for ref in refs}
-    for _slot_number, ref in _named_character_refs_for_text(rewritten, ref_list):
+    if include_named:
+        for _slot_number, ref in _named_refs_for_text(rewritten, ref_list, kinds=("character", "location")):
             marker = id(ref)
             if marker in seen:
                 continue
@@ -4606,6 +4635,30 @@ def resolve_prompt_refs(text, ref_list):
     return rewritten, refs, dropped
 
 
+def resolve_shot_references(text, ref_list, ref_mode="auto ref2v", shot_index=0, handoff=None):
+    """Compatibility wrapper for older tests and helper code.
+
+    The renderer now uses `resolve_prompt_refs` plus `shot_references` directly,
+    but some helper tests still check the combined resolution path."""
+    if ref_mode == "where tagged":
+        rewritten, refs, dropped = resolve_prompt_refs(text, ref_list, include_named=False)
+        return rewritten, refs, dropped, bool(picture_tags(text)), ref_mode
+    if ref_mode == "auto ref2v":
+        rewritten, tagged_refs, dropped = resolve_tagged_refs(text, ref_list)
+        refs = list(tagged_refs)
+        seen = {id(ref) for ref in refs}
+        for _slot_number, ref in _named_character_refs_for_text(rewritten, ref_list):
+            marker = id(ref)
+            if marker in seen:
+                continue
+            seen.add(marker)
+            refs.append(ref)
+        return rewritten, refs, dropped, bool(picture_tags(text)), ref_mode
+    rewritten, _, dropped = resolve_tagged_refs(text, ref_list)
+    refs = shot_references(ref_list, ref_mode, shot_index, handoff)
+    return rewritten, refs, dropped, False, ref_mode
+
+
 def shot_references(ref_list, ref_mode, shot_index, handoff):
     """Pure: which reference images shot `shot_index` is conditioned on, or [] when
     the shot should use the keyframe handoff instead.

tests/test_dumas_h3_longvideos.py

@@ -513,6 +513,20 @@ class DumasH3LongVideosHelperTests(unittest.TestCase):
                 ["img2", "img4", "img7"],
             )
 
+    def test_annotate_script_refs_handles_named_characters_and_locations(self):
+        refs = [
+            {"kind": "character", "image": "img1", "name": "Mara"},
+            {"kind": "location", "image": "img2", "name": "Hangar"},
+        ]
+
+        report = self.module.annotate_script_refs(
+            ["Mara waits in the Hangar.", "Nobody else is here."],
+            refs,
+        )
+
+        self.assertIn("# shot 1 refs: Picture 1 Mara (by name); Picture 2 Hangar (by name)", report)
+        self.assertIn("# shot 2 refs: none", report)
+
     def test_input_types_expose_nine_ref_slots(self):
         optional = self.module.H3LongVideos.INPUT_TYPES()["optional"]