Clean SDXL paired character tags

sdxl_tag_policy.py

@@ -64,6 +64,7 @@ def split_tag_text(text: Any) -> list[str]:
         flags=re.IGNORECASE,
     )
     text = re.sub(r"(?<!-)\b(?:and|with)\b(?!-)", ",", text, flags=re.IGNORECASE)
+    text = re.sub(r"\b(woman|man),\s+(woman|man)\s+are\b", r"\1 and \2 are", text, flags=re.IGNORECASE)
     parts = re.split(r"\s*[,;]\s*", text)
     return [clean(part).strip(" .") for part in parts if clean(part).strip(" .")]
 

tools/prompt_route_simulation.py

@@ -588,6 +588,12 @@ def _sdxl_hyphen_fragment_issues(name: str, sdxl_prompt: str) -> list[str]:
     return []
 
 
+def _sdxl_character_pair_fragment_issues(name: str, sdxl_prompt: str) -> list[str]:
+    if re.search(r"\b(?:woman,\s*man|man,\s*woman)\s+are\b", sdxl_prompt, flags=re.IGNORECASE):
+        return [f"{name}.sdxl_prompt: broken_character_pair_tag"]
+    return []
+
+
 def _trace_dict(formatter_name: str, payload: dict[str, Any]) -> tuple[dict[str, Any], str]:
     trace_text = str(payload.get("route_trace_json") or "")
     if not trace_text:
@@ -727,6 +733,7 @@ def _formatter_issues(
     issues.extend(_sdxl_composition_tag_issues(name, sdxl_prompt))
     issues.extend(_sdxl_expression_label_issues(name, sdxl_prompt))
     issues.extend(_sdxl_hyphen_fragment_issues(name, sdxl_prompt))
+    issues.extend(_sdxl_character_pair_fragment_issues(name, sdxl_prompt))
 
     for label, value in (
         (f"{name}.krea_negative", krea.get("negative_prompt")),

tools/prompt_smoke.py

@@ -4680,6 +4680,9 @@ def smoke_sdxl_tag_policy() -> None:
     _expect("front-and-back penetration" in hyphenated_tags, "SDXL tag splitter broke hyphenated and compound")
     _expect("front-" not in hyphenated_tags and "-back penetration" not in hyphenated_tags, "SDXL tag splitter emitted broken hyphen fragments")
     _expect("hands on hips" in hyphenated_tags, "SDXL tag splitter stopped splitting non-hyphenated with connector")
+    subject_pair_tags = sdxl_tag_policy.split_tag_text("Woman A, Man A are mid-transition with hands on hips")
+    _expect("woman and man are mid-transition" in subject_pair_tags, "SDXL tag splitter broke paired character clause")
+    _expect("woman" not in subject_pair_tags and "man are mid-transition" not in subject_pair_tags, "SDXL tag splitter emitted broken paired character fragments")
     _expect(
         sdxl_formatter._camera_tags(row) == sdxl_tag_policy.camera_tags(row),
         "SDXL formatter camera helper should delegate to sdxl_tag_policy",