From 2fae9208399cb8daa0ed51e74950caf3eb35efd2 Mon Sep 17 00:00:00 2001
From: digital-garbage <133010426+digital-garbage@users.noreply.github.com>
Date: Mon, 15 Jun 2026 12:39:52 +0300
Subject: [PATCH] =?UTF-8?q?Restore=20generic=20'scene=20N'=20split=20?=
 =?UTF-8?q?=E2=80=94=20but=20literal-only=20(Movie=20Editor=20needs=20it)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Dropping the generic scene-N pattern broke the Movie Editor: build_combined_prompt
(for_generation) INJECTS 'scene N' markers into the generation prompt to force a
split for any unit whose text lacks a leading transition trigger. Without that
recognition the second unit never split -> 'scene 2 exists only on the timeline,
not generated'.

split_scenes now matches the generic scene-N label again, but ONLY in its
literal-text scanner (never over shortcut expansions). So an editor-injected /
user-typed 'scene 2' cuts, while a 'scene 2' buried inside a shortcut expansion
does not — fixing both the user's phantom split AND the broken montage.

For the reported prompt the piece-walker now returns 2 scenes (was 4), matching
the old generation splitter's 2 -> the guard should pass and keys attribute while
the chain/guides keep working.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 conditioning.py               | 25 ++++++++++++++-----------
 tests/test_refinement_keys.py | 19 +++++++++++++++++++
 2 files changed, 33 insertions(+), 11 deletions(-)

diff --git a/conditioning.py b/conditioning.py
index be5ce56e..3f463c86 100644
--- a/conditioning.py
+++ b/conditioning.py
@@ -886,17 +886,20 @@ def split_scenes(prompt, placement="start"):
             _open(effect)
             segs[-1]["raw"] += raw_t; segs[-1]["exp"] += exp_t
 
-    # Literal-text trigger matcher (custom triggers only; never the generic scene-N pattern).
-    triggers = [t for t in custom_map.keys() if t]
-    lit_pat = None
-    if triggers:
-        def _tp(t):
-            end = r"\b" if re.search(r"\w$", t) else r"(?=\s|$)"
-            return r"\b" + re.escape(t) + end
-        try:
-            lit_pat = re.compile("|".join(_tp(t) for t in sorted(triggers, key=len, reverse=True)), re.IGNORECASE)
-        except re.error:
-            lit_pat = None
+    # Literal-text trigger matcher: the user's custom triggers PLUS the generic `scene <N>` label.
+    # Scoped to LITERAL text only (this matcher never runs over shortcut expansions), so a "scene 2"
+    # the user types — or, crucially, the one the Movie Editor INJECTS into the generation prompt to
+    # force a split for a scene with no leading trigger (see build_combined_prompt for_generation) —
+    # still cuts, while a "scene 2" buried inside a shortcut's expansion does NOT.
+    def _tp(t):
+        end = r"\b" if re.search(r"\w$", t) else r"(?=\s|$)"
+        return r"\b" + re.escape(t) + end
+    _pats = [_tp(t) for t in sorted((t for t in custom_map.keys() if t), key=len, reverse=True)]
+    _pats.append(_GENERIC_SCENE_LABEL_PATTERN)
+    try:
+        lit_pat = re.compile("|".join(_pats), re.IGNORECASE)
+    except re.error:
+        lit_pat = None
 
     for pc in pieces:
         raw = text[pc.get("orig_start", 0):pc.get("orig_end", 0)]
diff --git a/tests/test_refinement_keys.py b/tests/test_refinement_keys.py
index b7225aa7..9333b49f 100644
--- a/tests/test_refinement_keys.py
+++ b/tests/test_refinement_keys.py
@@ -245,6 +245,25 @@ def test_split_scenes_ignores_triggers_inside_expansions(monkeypatch):
     assert "Scene cut." not in out["scenes"][1]["expanded"]  # silent trigger dropped
 
 
+def test_split_scenes_literal_scene_label_still_splits(monkeypatch):
+    """The generic 'scene N' label MUST still split when it's LITERAL text — the Movie Editor
+    injects it into the generation prompt to force a split for a unit with no leading trigger
+    (build_combined_prompt for_generation). But the SAME label inside a shortcut expansion must NOT
+    split (that was the user's phantom 'scene 2'). One rule: literal cuts, expansion-content doesn't."""
+    db = _db({"qcut": (["qcut"], ["rapid cut"], ""),
+              "tth": (["tth"], ["a backflip"], ""),
+              "cof": (["cof"], ["sitting, scene 9 of many"], "cs")})  # 'scene 9' INSIDE expansion
+    monkeypatch.setattr(templates, "load_shortcut_db", lambda: db)
+    monkeypatch.setattr(templates, "load_custom_transition_triggers",
+                        lambda: {"qcut": {"placement": "start", "visual_effect": "none"}})
+    # explicit qcut leads scene 0; the editor-injected literal 'scene 2' leads scene 1
+    out = conditioning.split_scenes("CHAR a woman man qcut tth scene 2 cof")
+    assert len(out["scenes"]) == 2
+    assert out["scenes"][0]["keys"] == set()      # tth has no key
+    assert out["scenes"][1]["keys"] == {"cs"}     # cof, despite 'scene 9' in its expansion
+    assert out["anchor"] == "CHAR a woman man"
+
+
 def test_single_scene_with_custom_key_is_detected(monkeypatch):
     """Regression: a custom key fired in a single-logical-scene prompt MUST be detected (was
     silently dropped to empty when the old code compared scene counts and fell back)."""