auto-save 2026-05-14 05:05 (~6)

api/main.py

@@ -96,7 +96,7 @@ AssetBackground = Literal["white", "black"]
 AssetSize = Literal["source", "1024", "1536", "2048"]
 AssetQuality = Literal["hd"]
 SubjectKind = Literal["object", "living"]
-SubjectView = Literal["front", "back", "left", "right", "side", "side_walk", "top", "bottom", "expression"]
+SubjectView = str
 FRAME_TARGET_LABELS: dict[FrameExtractTarget, str] = {
     "balanced": "综合关键帧",
     "subject": "清晰主体",
@@ -690,16 +690,43 @@ def _make_reference_contact_sheet(job_id: str, frame_indices: list[int], out_pat
     return out_path
 
 
-def _subject_view_labels(kind: SubjectKind) -> list[tuple[SubjectView, str]]:
+SUBJECT_VIEW_LABELS: dict[str, str] = {
+    "front": "正面",
+    "back": "背面",
+    "left": "左侧",
+    "right": "右侧",
+    "side": "侧面",
+    "side_walk": "侧面走路",
+    "top": "顶部视角",
+    "bottom": "底部视角",
+    "expression_neutral": "中性表情",
+    "expression_happy": "开心表情",
+    "expression_angry": "生气表情",
+    "expression_sad": "难过表情",
+    "expression_relaxed": "放松表情",
+    "action_walk": "走路动作",
+    "action_sit": "坐姿动作",
+    "action_hold": "手持动作",
+    "action_use": "使用动作",
+}
+
+
+def _subject_view_labels(kind: SubjectKind, requested: list[str] | None = None) -> list[tuple[SubjectView, str]]:
+    if requested:
+        normalized: list[str] = []
+        for raw in requested:
+            key = "".join(ch for ch in str(raw).strip().lower() if ch.isalnum() or ch == "_")
+            if key and key not in normalized:
+                normalized.append(key)
+        return [(key, SUBJECT_VIEW_LABELS.get(key, key.replace("_", " "))) for key in normalized[:12]]
     if kind == "living":
         return [
             ("front", "正面站立"),
             ("back", "背面站立"),
             ("side", "侧面站立"),
             ("side_walk", "侧面走路"),
-            ("top", "顶部视角"),
-            ("bottom", "底部视角"),
-            ("expression", "表情参考"),
+            ("expression_neutral", "中性表情"),
+            ("expression_relaxed", "放松表情"),
         ]
     return [
         ("front", "正面"),
@@ -1911,6 +1938,7 @@ class GenerateSubjectAssetsReq(BaseModel):
     quality: AssetQuality = "hd"
     size: AssetSize = "source"
     source_frame_indices: list[int] | None = None
+    views: list[str] | None = None
 
 
 @app.post("/jobs/{job_id}/frames/{idx}/elements", response_model=Job)
@@ -2205,11 +2233,14 @@ def generate_subject_assets(job_id: str, idx: int, element_id: str, req: Generat
     models = [IMAGE_MODEL, "gemini-3.1-flash-image-preview", "gemini-2.5-flash-image"]
     generated: list[SubjectAsset] = []
     try:
-        for view, view_label in _subject_view_labels(req.subject_kind):
+        for view, view_label in _subject_view_labels(req.subject_kind, req.views):
             if view == "side_walk":
                 view_prompt = "side view in a natural walking pose, same identity and proportions"
-            elif view == "expression":
-                view_prompt = "clear expression reference, frontal or three-quarter standing pose, preserving the same identity"
+            elif view.startswith("expression_"):
+                emotion = view_label.replace("表情", "")
+                view_prompt = f"clear {emotion} facial expression reference, frontal or three-quarter standing pose, preserving the same identity"
+            elif view.startswith("action_"):
+                view_prompt = f"{view_label} reference pose, same identity and proportions"
             else:
                 view_prompt = f"{view_label} view"
             prompt = (