auto-save 2026-05-18 00:23 (~2)

2026-05-18 00:23:43 +08:00
parent eeff64c6e5
commit a7b131534a
2 changed files with 137 additions and 37 deletions
--- a/.memory/worklog.json
+++ b/.memory/worklog.json
@@ -1,25 +1,5 @@
 {
  "entries": [
-    {
-      "files_changed": 10,
-      "hash": "f7cc49a",
-      "message": "auto-save 2026-05-15 15:21 (+1, ~9)",
-      "ts": "2026-05-15T15:21:20+08:00",
-      "type": "commit"
-    },
-    {
-      "files_changed": 2,
-      "message": "Codex 会话活跃 · 最近命令：codex · 2 项未提交变更 · 最近提交：auto-save 2026-05-15 15:21 (+1, ~9)",
-      "ts": "2026-05-15T07:24:47Z",
-      "type": "session-heartbeat"
-    },
-    {
-      "files_changed": 3,
-      "hash": "caa28e2",
-      "message": "auto-save 2026-05-15 15:26 (~3)",
-      "ts": "2026-05-15T15:26:51+08:00",
-      "type": "commit"
-    },
    {
      "files_changed": 1,
      "hash": "45e7401",
@@ -3258,6 +3238,26 @@
      "type": "session-heartbeat",
      "message": "Codex 会话活跃 · 最近命令：codex · 分支 main · 2 项未提交变更 · 最近提交：auto-save 2026-05-18 00:07 (~3)",
      "files_changed": 2
+    },
+    {
+      "ts": "2026-05-18T00:12:58+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-18 00:12 (~3)",
+      "hash": "ba202e4",
+      "files_changed": 3
+    },
+    {
+      "ts": "2026-05-18T00:16:10+08:00",
+      "type": "commit",
+      "message": "fix: show generated subject views",
+      "hash": "eeff64c",
+      "files_changed": 1
+    },
+    {
+      "ts": "2026-05-17T16:18:31Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 分支 main · 1 项未提交变更 · 最近提交：fix: show generated subject views",
+      "files_changed": 1
    }
  ]
 }
--- a/api/main.py
+++ b/api/main.py
@@ -50,8 +50,10 @@ LOCAL_ASR_TIMEOUT_SECONDS = max(30, int(os.getenv("LOCAL_ASR_TIMEOUT_SECONDS", "
 TRANSLATE_MODEL = os.getenv("TRANSLATE_MODEL", "gemini-2.5-flash")
 REWRITE_MODEL = os.getenv("REWRITE_MODEL", "gemini-2.5-pro")
 VISION_MODEL = os.getenv("VISION_MODEL", "gemini-2.5-flash")
-IMAGE_MODEL = os.getenv("IMAGE_MODEL", "gemini-3-pro-image-preview")
 GPT_IMAGE_MODEL = os.getenv("GPT_IMAGE_MODEL", "gpt-image-2").strip() or "gpt-image-2"
+IMAGE_BASE_URL = os.getenv("IMAGE_BASE_URL", LLM_BASE_URL).strip()
+IMAGE_API_KEY = os.getenv("IMAGE_API_KEY", LLM_API_KEY).strip()
+IMAGE_MODEL = os.getenv("IMAGE_MODEL", GPT_IMAGE_MODEL).strip() or GPT_IMAGE_MODEL
 SUBJECT_ASSET_IMAGE_MODEL = os.getenv("SUBJECT_ASSET_IMAGE_MODEL", GPT_IMAGE_MODEL).strip() or GPT_IMAGE_MODEL
 SUBJECT_ASSET_IMAGE_MODELS = [
    m.strip()
@@ -87,6 +89,18 @@ MINIMAX_TTS_VOICE_POOL = [
    for v in os.getenv("MINIMAX_TTS_VOICE_POOL", ",".join(DEFAULT_MINIMAX_TTS_VOICE_POOL)).split(",")
    if v.strip()
 ]
+VOICE_PROVIDER = os.getenv("VOICE_PROVIDER", "azure_openai").strip().lower() or "azure_openai"
+AZURE_OPENAI_BASE_URL = os.getenv("AZURE_OPENAI_BASE_URL", "https://ai.skg.com/azure").strip().rstrip("/")
+AZURE_OPENAI_API_KEY = os.getenv("AZURE_OPENAI_API_KEY", LLM_API_KEY).strip()
+AZURE_TTS_MODEL = os.getenv("AZURE_TTS_MODEL", "gpt-4o-mini-tts").strip() or "gpt-4o-mini-tts"
+AZURE_TTS_VOICE_ID = os.getenv("AZURE_TTS_VOICE_ID", "alloy").strip() or "alloy"
+DEFAULT_AZURE_TTS_VOICE_POOL = ["alloy", "verse", "shimmer"]
+AZURE_TTS_VOICE_POOL = [
+    v.strip()
+    for v in os.getenv("AZURE_TTS_VOICE_POOL", ",".join(DEFAULT_AZURE_TTS_VOICE_POOL)).split(",")
+    if v.strip()
+]
+AZURE_TTS_PATH = os.getenv("AZURE_TTS_PATH", "/audio/speech").strip() or "/audio/speech"

 POE_API_BASE_URL = os.getenv("POE_API_BASE_URL", "https://api.poe.com/v1").strip() or "https://api.poe.com/v1"
 POE_API_KEY = os.getenv("POE_API_KEY", "").strip()
@@ -163,6 +177,7 @@ _MEDIA_BIN_CACHE: dict[str, str] = {}
 # OpenAI 客户端（OpenAI 兼容网关，含 SKG ezlink）
 from openai import OpenAI
 _llm_client: OpenAI | None = None
+_image_client: OpenAI | None = None
 def llm() -> OpenAI:
    global _llm_client
    if _llm_client is None:
@@ -171,6 +186,14 @@ def llm() -> OpenAI:
        _llm_client = OpenAI(base_url=LLM_BASE_URL or None, api_key=LLM_API_KEY)
    return _llm_client

+def image_llm() -> OpenAI:
+    global _image_client
+    if _image_client is None:
+        if not IMAGE_API_KEY:
+            raise RuntimeError("IMAGE_API_KEY 或 LLM_API_KEY 未配置")
+        _image_client = OpenAI(base_url=IMAGE_BASE_URL or None, api_key=IMAGE_API_KEY)
+    return _image_client
+
 # Pipeline 状态：
 #   created → downloading → downloaded（前端“开始”会继续触发音频解析）
 #         → splitting → frames_extracted
@@ -2180,6 +2203,18 @@ def _choose_minimax_voice_id() -> str:
    return MINIMAX_TTS_VOICE_ID


+def _choose_azure_voice_id() -> str:
+    if AZURE_TTS_VOICE_POOL:
+        return random.choice(AZURE_TTS_VOICE_POOL)
+    return AZURE_TTS_VOICE_ID
+
+
+def _choose_tts_voice_id() -> str:
+    if VOICE_PROVIDER == "azure_openai":
+        return _choose_azure_voice_id()
+    return _choose_minimax_voice_id()
+
+
 def _voice_speed_for(voice_id: str, target_seconds: float, text: str) -> float:
    words = len([w for w in text.replace("\n", " ").split(" ") if w.strip()])
    estimated_seconds = words / 2.35 if words else target_seconds
@@ -2241,17 +2276,71 @@ def _minimax_tts_sync(job_id: str, text: str, voice_id: str, target_seconds: flo
    return f"/jobs/{job_id}/audio-script.mp3"


+def _azure_tts_url() -> str:
+    path = AZURE_TTS_PATH if AZURE_TTS_PATH.startswith("/") else f"/{AZURE_TTS_PATH}"
+    if AZURE_OPENAI_BASE_URL.endswith(path):
+        return AZURE_OPENAI_BASE_URL
+    return f"{AZURE_OPENAI_BASE_URL}{path}"
+
+
+def _azure_openai_tts_sync(job_id: str, text: str, voice_id: str, target_seconds: float = 12.0) -> str:
+    if not AZURE_OPENAI_API_KEY:
+        raise RuntimeError("AZURE_OPENAI_API_KEY 或 LLM_API_KEY 未配置，未生成配音")
+    if not text.strip():
+        raise RuntimeError("改写文案为空，未生成配音")
+    payload = {
+        "model": AZURE_TTS_MODEL,
+        "voice": voice_id,
+        "input": text.strip()[:9500],
+        "response_format": "mp3",
+        "speed": _voice_speed_for(voice_id, target_seconds, text),
+    }
+    resp = httpx.post(
+        _azure_tts_url(),
+        headers={
+            "Authorization": f"Bearer {AZURE_OPENAI_API_KEY}",
+            "api-key": AZURE_OPENAI_API_KEY,
+            "Content-Type": "application/json",
+        },
+        json=payload,
+        timeout=120,
+    )
+    if resp.status_code >= 400:
+        raise RuntimeError(f"Azure OpenAI TTS HTTP {resp.status_code}: {resp.text[:300]}")
+    audio_bytes = resp.content
+    if not audio_bytes:
+        raise RuntimeError("Azure OpenAI TTS 未返回音频内容")
+    content_type = resp.headers.get("content-type", "")
+    if "application/json" in content_type.lower():
+        try:
+            data = resp.json()
+        except Exception:
+            data = {"error": resp.text[:300]}
+        raise RuntimeError(f"Azure OpenAI TTS 返回 JSON 而不是音频：{str(data)[:300]}")
+    out = job_dir(job_id) / "audio_script.mp3"
+    out.write_bytes(audio_bytes)
+    return f"/jobs/{job_id}/audio-script.mp3"
+
+
+def _tts_sync(job_id: str, text: str, voice_id: str, target_seconds: float = 12.0) -> tuple[str, str, str]:
+    if VOICE_PROVIDER == "azure_openai":
+        return _azure_openai_tts_sync(job_id, text, voice_id, target_seconds), "azure_openai", AZURE_TTS_MODEL
+    return _minimax_tts_sync(job_id, text, voice_id, target_seconds), "minimax", MINIMAX_TTS_MODEL
+
+
 def _build_audio_script_sync(job_id: str, segments: list[TranscriptSegment], target_seconds: float = 12.0) -> AudioScript:
    source_text = _transcript_join(segments, "en")
    source_zh = _transcript_join(segments, "zh")
    duration = max(float(target_seconds or 0), _segment_duration(segments), 4.0)
    rewritten, rewrite_error = _rewrite_audio_script_sync(segments, duration)
-    selected_voice_id = _choose_minimax_voice_id()
+    selected_voice_id = _choose_tts_voice_id()
    speaker_profile, rhythm_profile = _audio_delivery_profile(segments, duration, selected_voice_id)
    voice_url = ""
    voice_error = ""
+    voice_provider = "azure_openai" if VOICE_PROVIDER == "azure_openai" else "minimax"
+    voice_model = AZURE_TTS_MODEL if voice_provider == "azure_openai" else MINIMAX_TTS_MODEL
    try:
-        voice_url = _minimax_tts_sync(job_id, rewritten, selected_voice_id, duration)
+        voice_url, voice_provider, voice_model = _tts_sync(job_id, rewritten, selected_voice_id, duration)
    except Exception as e:
        voice_error = str(e)
    # 改写失败时已有本地 SKG 模板兜底，不把它标成用户可见错误；配音失败才需要提示。
@@ -2265,8 +2354,8 @@ def _build_audio_script_sync(job_id: str, segments: list[TranscriptSegment], tar
        rhythm_profile=rhythm_profile,
        product_brief=AUDIO_PRODUCT_BRIEF,
        rewrite_model=AUDIO_REWRITE_MODEL,
-        voice_provider="minimax",
-        voice_model=MINIMAX_TTS_MODEL,
+        voice_provider=voice_provider,
+        voice_model=voice_model,
        voice_id=selected_voice_id,
        voice_url=voice_url,
        error=errors,
@@ -2453,8 +2542,8 @@ def _image_edit_call(
    import time as _time
    import httpx
    from PIL import Image as _PILImage
-    if not LLM_API_KEY:
-        raise RuntimeError("LLM_API_KEY 未配置")
+    if not IMAGE_API_KEY:
+        raise RuntimeError("IMAGE_API_KEY 或 LLM_API_KEY 未配置")
    # model 优先级：models 列表 > 单个 model 参数 > IMAGE_MODEL
    if models and len(models) > 0:
        models_cycle = list(models)
@@ -2489,9 +2578,9 @@ def _image_edit_call(
            if current_mode == "edit":
                with httpx.Client(timeout=120) as client:
                    r = client.post(
-                        f"{LLM_BASE_URL}/images/generations",
+                        f"{IMAGE_BASE_URL}/images/generations",
                        headers={
-                            "Authorization": f"Bearer {LLM_API_KEY}",
+                            "Authorization": f"Bearer {IMAGE_API_KEY}",
                            "Content-Type": "application/json",
                        },
                        json={"model": current_model, "prompt": prompt, "image": data_uri, "n": 1},
@@ -2499,7 +2588,7 @@ def _image_edit_call(
                    r.raise_for_status()
                    resp_data = r.json()
            else:
-                resp = llm().images.generate(model=current_model, prompt=prompt, n=1)
+                resp = image_llm().images.generate(model=current_model, prompt=prompt, n=1)
                resp_data = resp.model_dump() if hasattr(resp, "model_dump") else {"data": [{"b64_json": resp.data[0].b64_json}]}
            if resp_data.get("data"):
                effective_mode = current_mode
@@ -2542,15 +2631,15 @@ def _image_text_call(
    """Text-only image generation with light model rotation."""
    import base64 as b64lib
    import time as _time
-    if not LLM_API_KEY:
-        raise RuntimeError("LLM_API_KEY 未配置")
+    if not IMAGE_API_KEY:
+        raise RuntimeError("IMAGE_API_KEY 或 LLM_API_KEY 未配置")
    models_cycle = list(models) if models else [model or IMAGE_MODEL]
    last_err = ""
    resp_data: dict = {}
    for attempt in range(max_attempts):
        current_model = models_cycle[min(attempt, len(models_cycle) - 1)]
        try:
-            resp = llm().images.generate(model=current_model, prompt=prompt, n=1)
+            resp = image_llm().images.generate(model=current_model, prompt=prompt, n=1)
            resp_data = resp.model_dump() if hasattr(resp, "model_dump") else {"data": [{"b64_json": resp.data[0].b64_json}]}
            if resp_data.get("data"):
                b64 = resp_data["data"][0].get("b64_json")
@@ -2752,6 +2841,8 @@ def health() -> dict:
        "llm_configured": bool(LLM_API_KEY),
        "auth_configured": WEB_AUTH_CONFIGURED,
        "base_url": LLM_BASE_URL or "openai-default",
+        "image_base_url": IMAGE_BASE_URL or LLM_BASE_URL or "openai-default",
+        "voice_base_url": AZURE_OPENAI_BASE_URL if VOICE_PROVIDER == "azure_openai" else MINIMAX_TTS_BASE_URL,
        "models": {
            "asr": ASR_MODEL,
            "local_asr": LOCAL_ASR_MODEL,
@@ -2761,9 +2852,16 @@ def health() -> dict:
            "audio_rewrite": AUDIO_REWRITE_MODEL,
            "vision": VISION_MODEL,
            "image": IMAGE_MODEL,
-            "image_fallbacks": [IMAGE_MODEL, "gemini-3.1-flash-image-preview", "gemini-2.5-flash-image"],
+            "image_base_url": IMAGE_BASE_URL or LLM_BASE_URL or "openai-default",
+            "image_fallbacks": [IMAGE_MODEL, GPT_IMAGE_MODEL, "gpt-image-1.5"],
            "subject_image": SUBJECT_ASSET_IMAGE_MODEL,
            "subject_image_fallbacks": SUBJECT_ASSET_IMAGE_MODELS,
+            "voice_provider": VOICE_PROVIDER,
+            "voice_base_url": AZURE_OPENAI_BASE_URL if VOICE_PROVIDER == "azure_openai" else MINIMAX_TTS_BASE_URL,
+            "voice_tts": AZURE_TTS_MODEL if VOICE_PROVIDER == "azure_openai" else MINIMAX_TTS_MODEL,
+            "voice_id": AZURE_TTS_VOICE_ID if VOICE_PROVIDER == "azure_openai" else MINIMAX_TTS_VOICE_ID,
+            "voice_pool": AZURE_TTS_VOICE_POOL if VOICE_PROVIDER == "azure_openai" else (MINIMAX_TTS_VOICE_POOL or [MINIMAX_TTS_VOICE_ID]),
+            "voice_configured": bool(AZURE_OPENAI_API_KEY) if VOICE_PROVIDER == "azure_openai" else bool(MINIMAX_API_KEY),
            "minimax_tts": MINIMAX_TTS_MODEL,
            "minimax_voice": MINIMAX_TTS_VOICE_ID,
            "minimax_voice_pool": MINIMAX_TTS_VOICE_POOL or [MINIMAX_TTS_VOICE_ID],
@@ -3049,6 +3147,8 @@ def generate_image(job_id: str, idx: int, req: GenerateReq) -> Job:
        full_prompt = f"{full_prompt}. Avoid: {req.negative_prompt.strip()}"
    if not full_prompt:
        raise HTTPException(400, "prompt required")
+    if not IMAGE_API_KEY:
+        raise HTTPException(503, "IMAGE_API_KEY 或 LLM_API_KEY 未配置")

    model = req.model or IMAGE_MODEL
    gen_id = uuid.uuid4().hex[:12]
@@ -3075,9 +3175,9 @@ def generate_image(job_id: str, idx: int, req: GenerateReq) -> Job:
                # OpenAI SDK 不直接支持 image 参数，用底层 httpx
                with httpx.Client(timeout=120) as client:
                    r = client.post(
-                        f"{LLM_BASE_URL}/images/generations",
+                        f"{IMAGE_BASE_URL}/images/generations",
                        headers={
-                            "Authorization": f"Bearer {LLM_API_KEY}",
+                            "Authorization": f"Bearer {IMAGE_API_KEY}",
                            "Content-Type": "application/json",
                        },
                        json={
@@ -3091,7 +3191,7 @@ def generate_image(job_id: str, idx: int, req: GenerateReq) -> Job:
                    resp_data = r.json()
            else:
                # text-only
-                resp = llm().images.generate(model=model, prompt=full_prompt, n=1)
+                resp = image_llm().images.generate(model=model, prompt=full_prompt, n=1)
                resp_data = resp.model_dump() if hasattr(resp, "model_dump") else {"data": [{"b64_json": resp.data[0].b64_json}]}

            if resp_data.get("data"):