fix: normalize media prompts and patent views

api/main.py

@@ -2664,6 +2664,15 @@ SUBJECT_VIEW_LABELS: dict[str, str] = {
     "back_detail": "背部特写",
 }
 
+OBJECT_PATENT_VIEW_LABELS: dict[str, str] = {
+    "front": "正投影主视图",
+    "back": "正投影后视图",
+    "left": "正投影左视图",
+    "right": "正投影右视图",
+    "top": "正投影俯视图",
+    "bottom": "正投影仰视图",
+}
+
 
 def _subject_view_labels(kind: SubjectKind, requested: list[str] | None = None) -> list[tuple[SubjectView, str]]:
     if requested:
@@ -2672,7 +2681,8 @@ def _subject_view_labels(kind: SubjectKind, requested: list[str] | None = None)
             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[:10]]
+        labels = OBJECT_PATENT_VIEW_LABELS if kind == "object" else SUBJECT_VIEW_LABELS
+        return [(key, labels.get(key, SUBJECT_VIEW_LABELS.get(key, key.replace("_", " ")))) for key in normalized[:10]]
     if kind == "living":
         return [
             ("front", "正面站立"),
@@ -2687,16 +2697,36 @@ def _subject_view_labels(kind: SubjectKind, requested: list[str] | None = None)
             ("back_neck_detail", "后颈/肩背特写"),
         ]
     return [
-        ("front", "正面"),
-        ("back", "背面"),
-        ("left", "左侧"),
-        ("right", "右侧"),
-        ("top", "正投影俯视图"),
-        ("bottom", "正投影仰视图"),
+        ("front", OBJECT_PATENT_VIEW_LABELS["front"]),
+        ("back", OBJECT_PATENT_VIEW_LABELS["back"]),
+        ("left", OBJECT_PATENT_VIEW_LABELS["left"]),
+        ("right", OBJECT_PATENT_VIEW_LABELS["right"]),
+        ("top", OBJECT_PATENT_VIEW_LABELS["top"]),
+        ("bottom", OBJECT_PATENT_VIEW_LABELS["bottom"]),
     ]
 
 
 def _subject_view_projection_clause(view: str) -> str:
+    if view == "front":
+        return (
+            "Patent-style orthographic main/front elevation view: look straight at the designated main face, "
+            "with the viewing direction perpendicular to that face. No perspective, no tilt, no 3/4 angle, no isometric view. "
+        )
+    if view == "back":
+        return (
+            "Patent-style orthographic rear elevation view: look straight at the rear face, "
+            "with the viewing direction perpendicular to that face. No perspective, no tilt, no 3/4 angle, no isometric view. "
+        )
+    if view == "left":
+        return (
+            "Patent-style orthographic left side elevation view: look straight at the product's left side, "
+            "with the viewing direction perpendicular to that side face. No perspective, no tilt, no 3/4 angle, no isometric view. "
+        )
+    if view == "right":
+        return (
+            "Patent-style orthographic right side elevation view: look straight at the product's right side, "
+            "with the viewing direction perpendicular to that side face. No perspective, no tilt, no 3/4 angle, no isometric view. "
+        )
     if view == "top":
         return (
             "Patent-style orthographic top view: look straight down from directly above the product, "
@@ -5597,13 +5627,14 @@ def generate_image(job_id: str, idx: int, req: GenerateReq) -> Job:
                 reference_path = sel_path
                 reference_source = f"gen:{sel.id[:6]}"
 
-    full_prompt = req.prompt.strip()
+    raw_prompt = req.prompt.strip()
     if req.extra_prompt.strip():
-        full_prompt = f"{full_prompt}. Include: {req.extra_prompt.strip()}"
+        raw_prompt = f"{raw_prompt}. Include: {req.extra_prompt.strip()}"
     if req.negative_prompt.strip():
-        full_prompt = f"{full_prompt}. Avoid: {req.negative_prompt.strip()}"
-    if not full_prompt:
+        raw_prompt = f"{raw_prompt}. Avoid: {req.negative_prompt.strip()}"
+    if not raw_prompt:
         raise HTTPException(400, "prompt required")
+    full_prompt = _ensure_english(raw_prompt)
     if not IMAGE_API_KEY:
         raise HTTPException(503, "IMAGE_API_KEY 或 LLM_API_KEY 未配置")
 

docs/source-analysis.html

@@ -1181,6 +1181,19 @@ ProductRefStateItem {
         <h2>变更记录</h2>
         <p>这个记录不是 git log 的替代品。它记录“产品理解发生了什么变化、影响了哪些源码、你以后描述需求时该怎么说”。后续每次改功能都要补一条。</p>
         <div class="changelog">
+          <article class="change">
+            <header>
+              <h3>2026-05-25 · 生图生视频统一英文提示词和完整正投影视图</h3>
+              <span class="tag amber">API</span>
+              <span class="tag violet">UI</span>
+              <span class="tag blue">Docs</span>
+            </header>
+            <div class="body">
+              <p><strong>问题：</strong>生视频链路已经会把中文要求转成英文再提交外部视频 API，但首页直接生图仍可能把中文混合 prompt 直接传给图片 API；物体六视图也只对俯视/仰视加了正投影约束，前后左右仍有“正面/背面/左侧/右侧”的口语化表述。</p>
+              <p><strong>改动：</strong><code>/jobs/{job_id}/frames/{idx}/generate</code> 在调用图片 API 前统一对完整 prompt 执行 <code>_ensure_english()</code>，和生视频 <code>_enqueue_storyboard_videos</code> 的英文提交规则保持一致。物体主体套图的默认六面视图改为 <code>正投影主视图 / 正投影后视图 / 正投影左视图 / 正投影右视图 / 正投影俯视图 / 正投影仰视图</code>，并对前后左右也加入 <code>orthographic elevation</code>、<code>no perspective</code>、<code>no tilt</code>、<code>no 3/4 angle</code>、<code>no isometric view</code> 约束。</p>
+              <p><strong>影响：</strong>用户仍可以用中文描述生图或生视频需求；后端实际提交给图片/视频模型的提示词会优先归一为英文。以后“专利六视图”应明确理解为六面正投影视图，不等同于摄影角度图；真人/短视频主体参考包仍保留口播、45 度、肩颈近景等非专利视角。</p>
+            </div>
+          </article>
           <article class="change">
             <header>
               <h3>2026-05-25 · 主体套图俯视和仰视改为正投影口径</h3>

web/components/lightbox.tsx

@@ -32,10 +32,10 @@ interface Props {
 }
 
 const OBJECT_VIEW_OPTIONS = [
-  ["front", "正面"],
-  ["back", "背面"],
-  ["left", "左侧"],
-  ["right", "右侧"],
+  ["front", "正投影主视图"],
+  ["back", "正投影后视图"],
+  ["left", "正投影左视图"],
+  ["right", "正投影右视图"],
   ["top", "正投影俯视图"],
   ["bottom", "正投影仰视图"],
 ]