feat: 插件开发 ozon 端主体完成

server/services/generator.py

@@ -23,6 +23,18 @@ from services.prompt import build_prompt, build_context, type_name
 
 log = logging.getLogger("suite.generator")
 
+
+def _raise_api_error(resp, provider: str):
+    """HTTP 错误时抛出带 API 错误码/信息的异常（响应体里有真正的失败原因）。"""
+    if resp.is_success:
+        return
+    try:
+        body = resp.json()
+        detail = f"{body.get('code', '')}: {body.get('message', '')}".strip(': ')
+    except Exception:  # noqa: BLE001
+        detail = resp.text[:200]
+    raise RuntimeError(f"{provider} API HTTP {resp.status_code} — {detail or '无错误详情'}")
+
 # 参考图选择：material 用第 2 张（背面/细节），其余用第 1 张（正面）
 TYPE_REF_INDEX = {
     "material": 1,
@@ -81,12 +93,12 @@ async def _resolve_ref(url: str) -> str:
 
 # ── Provider：豆包 Seedream（火山方舟）────────────────────────────────────
 
-async def generate_doubao(prompt: str, ref_images: list[str], size: str = "2048x2048") -> bytes:
+async def generate_doubao(prompt: str, ref_images: list[str], size: str = "2048x2048", model: str | None = None) -> bytes:
     s = get_settings()
     if not s.ark_api_key:
         raise RuntimeError("未配置 ARK_API_KEY（.env）")
     body = {
-        "model": s.ark_image_model,
+        "model": model or s.ark_image_model,
         "prompt": prompt.rstrip(". ") + ". " + _DOUBAO_ANTI_AI,
         "size": size,
         "response_format": "url",
@@ -101,7 +113,7 @@ async def generate_doubao(prompt: str, ref_images: list[str], size: str = "2048x
             headers={"Authorization": f"Bearer {s.ark_api_key}", "Content-Type": "application/json"},
             json=body,
         )
-        resp.raise_for_status()
+        _raise_api_error(resp, "豆包")
         img_url = resp.json()["data"][0]["url"]
         dl = await client.get(img_url, timeout=s.request_timeout)
         dl.raise_for_status()
@@ -140,11 +152,12 @@ async def _tongyi_poll_task(client: httpx.AsyncClient, key: str, task_id: str, m
     raise TimeoutError(f"通义异步任务超时 ({max_wait}s): task_id={task_id}")
 
 
-async def generate_tongyi(prompt: str, ref_images: list[str], size: str = "2048*2048") -> bytes:
+async def generate_tongyi(prompt: str, ref_images: list[str], size: str = "2048*2048", model: str | None = None) -> bytes:
     s = get_settings()
     if not s.dashscope_api_key:
         raise RuntimeError("未配置 DASHSCOPE_API_KEY（.env）")
-    is_wan = _is_wan_model(s.dashscope_model)
+    model = model or s.dashscope_model
+    is_wan = _is_wan_model(model)
     url = s.dashscope_base_url or (
         "https://dashscope.aliyuncs.com/api/v1/services/aigc/image-generation/generation"
         if is_wan
@@ -163,11 +176,11 @@ async def generate_tongyi(prompt: str, ref_images: list[str], size: str = "2048*
     if is_wan:
         headers["X-DashScope-Async"] = "enable"
 
-    body = {"model": s.dashscope_model, "input": {"messages": [{"role": "user", "content": content}]}, "parameters": params}
+    body = {"model": model, "input": {"messages": [{"role": "user", "content": content}]}, "parameters": params}
 
     async with httpx.AsyncClient(timeout=s.request_timeout, verify=False) as client:
         resp = await client.post(url, headers=headers, json=body)
-        resp.raise_for_status()
+        _raise_api_error(resp, "通义")
         data = resp.json()
         if is_wan:
             task_id = data.get("output", {}).get("task_id", "")
@@ -267,7 +280,11 @@ async def run_suite(suite_id: str) -> None:
 
         raw = suite.context if not product else (product.raw or {})
         ctx = build_context(raw or {}, fallback_name=product.name if product else "")
-        size = _image_size(provider_name, suite.ratio, is_wan=_is_wan_model(settings.dashscope_model))
+        model = suite.model or (
+            settings.dashscope_model if provider_name == "tongyi" else settings.ark_image_model
+        )
+        is_wan = provider_name == "tongyi" and _is_wan_model(model)
+        size = _image_size(provider_name, suite.ratio, is_wan=is_wan)
 
         # 任务列表：方案（逐张）优先，旧路径按 types
         if suite.plan:
@@ -290,12 +307,15 @@ async def run_suite(suite_id: str) -> None:
             db.add(image_row)
             await db.flush()
             try:
-                prompt = build_prompt(type_id, ctx, suite.style_set, suite.lang, extra=job)
+                prompt = build_prompt(
+                    type_id, ctx, suite.style_set, suite.lang,
+                    extra=job, style_prompt=suite.style_prompt,
+                )
                 if product:
                     refs = await _select_ref_images(db, product.id, type_id)
                 else:
                     refs = _refs_for_job(list(suite.ref_images or []), job)
-                data = await generator(prompt, refs, size=size)
+                data = await generator(prompt, refs, size=size, model=model)
                 key = storage.write_bytes(data, key_prefix=f"suites/{suite.id}", ext=".jpg")
                 image_row.stored_url = storage.public_url(key)
                 image_row.status = STATUS_OK

server/services/planner.py

@@ -7,6 +7,7 @@ from __future__ import annotations
 
 import json
 import logging
+import re
 
 import httpx
 
@@ -21,29 +22,26 @@ ALLOWED_KINDS = [
     "size_chart", "sku_collection", "custom",
 ]
 
-SYSTEM_PROMPT = """你是一名资深电商视觉策划。根据商品信息规划一套电商详情页/主图套图的出图方案。
+SYSTEM_PROMPT = """你是一名资深电商视觉策划。根据商品信息规划一套电商套图的出图方案。
+
+## 输出硬性约束（违反即失败）
+1. 输出必须是**单行紧凑 JSON**：无换行、无缩进、无空格填充、无注释、无 markdown 围栏。
+2. 顶层只有 summary 和 items 两个字段；每个 item 严格只有 kind/title/detail/prompt_hint/count/variant_name 六个字段，不得增删。
+3. 文本长度上限（中文字符/英文单词数）：summary ≤ 25 字；title ≤ 8 字；detail ≤ 20 字；prompt_hint ≤ 15 个英文词。超限必须删减，不得省略号截断。
+4. count 默认 1，仅当该类图确需多个变体时才 >1，最大 3。总张数 8-15。
+5. variant_name 只能从「SKU规格」列表原样照抄；没有绑定就输出 null。
 
 ## 规划规则
-1. SKU 主图：商品有多个带图 SKU（颜色/款式）时，每个 SKU 出 1 张独立主图（kind=white_bg），
-   并在 variant_name 里填对应的 SKU 规格名（必须来自「SKU规格」列表，原样照抄）；
-   单 SKU 商品出 1 张主图即可（variant_name 留空）。
-2. 场景图（kind=lifestyle）：按商品的核心使用场景出 2-4 张，每张聚焦一个场景，场景从描述/参数里提取。
-3. 细节图（kind=material 或 custom）：按商品的关键细节/材质/结构出 2-3 张，每张聚焦一个卖点细节。
-4. 尺寸标注图（kind=size_chart）：参数里有长宽高/尺寸数据时出 1 张。
-5. SKU 合集图（kind=sku_collection）：SKU 数量 >1 时出 1 张，同款多色整齐排列。
-6. 可用 kind 枚举：white_bg / key_features / selling_pt / material / lifestyle / multi_scene /
-   ecommerce_detail / size_chart / sku_collection / custom。其他创意图用 custom。
-7. 总张数控制在 8-15 张；每项 count 为 1-3。
-8. title 用中文短语（≤8字，如「主图·粉色」「浴室壁挂场景」）；detail 用中文说明这张图要展示什么（≤40字）；
-   prompt_hint 用英文描述构图（角度/布局/光线要点，≤60 words），供生图模型使用。
+1. SKU 主图：每个带图 SKU 出 1 张独立主图（kind=white_bg），variant_name 填对应规格名；单 SKU 出 1 张（variant_name=null）。
+2. 场景图（kind=lifestyle）：按核心使用场景出 2-4 张，每张聚焦一个场景。
+3. 细节图（kind=material 或 custom）：按关键细节/材质/结构出 2-3 张，每张聚焦一个卖点。
+4. 尺寸标注图（kind=size_chart）：参数含长宽高/尺寸时出 1 张。
+5. SKU 合集图（kind=sku_collection）：SKU >1 时出 1 张。
+6. kind 枚举：white_bg / key_features / selling_pt / material / lifestyle / multi_scene / ecommerce_detail / size_chart / sku_collection / custom。
+7. title 用中文短语（如「主图·粉色」「浴室壁挂」）；detail 中文说明这张图展示什么；prompt_hint 用英文描述构图要点。
 
-## 输出格式（严格 JSON，不要多余文字）
-{
-  "summary": "整体思路一句话",
-  "items": [
-    {"kind": "white_bg", "title": "主图·粉色", "detail": "粉色SKU白底主视觉", "prompt_hint": "front view on pure white background", "count": 1, "variant_name": "粉色"}
-  ]
-}"""
+## 输出示例（紧凑单行）
+{"summary":"三色收纳盒全套图","items":[{"kind":"white_bg","title":"主图·粉色","detail":"粉色SKU白底主视觉","prompt_hint":"front view on white background","count":1,"variant_name":"粉色"}]}"""
 
 
 def _normalize_items(raw_items: list, sku_variants: list[str]) -> list[dict]:
@@ -76,6 +74,54 @@ def _normalize_items(raw_items: list, sku_variants: list[str]) -> list[dict]:
     return items
 
 
+def _repair_truncated(s: str) -> dict | None:
+    """截断修复：从最后一个完整的 '}' 处截断，剥尾逗号后按括号配平补全闭合。
+
+    适用于「items 数组中途被 max_tokens 截断」的场景——截断点在完整对象边界，
+    此前的字符串必然已闭合，简单计数配平即可。
+    """
+    for cut in (m.end() for m in reversed(list(re.finditer(r'\}', s)))):
+        cand = s[:cut].rstrip().rstrip(',')
+        opens: list[str] = []
+        for ch in cand:
+            if ch in '{[':
+                opens.append(ch)
+            elif ch == '}' and opens and opens[-1] == '{':
+                opens.pop()
+            elif ch == ']' and opens and opens[-1] == '[':
+                opens.pop()
+        suffix = ''.join('}' if o == '{' else ']' for o in reversed(opens))
+        try:
+            data = json.loads(cand + suffix)
+            if isinstance(data, dict):
+                return data
+        except json.JSONDecodeError:
+            continue
+    return None
+
+
+def _extract_json(text: str) -> dict:
+    """从模型输出提取 JSON：剥离思考块/markdown 围栏，截断时尝试修复。"""
+    s = (text or '').strip()
+    # 剥离思考块（思考型模型会把推理过程放进 <think>）
+    s = re.sub(r'<think>.*?</think>', '', s, flags=re.S).strip()
+    # 剥离 markdown 代码围栏
+    m = re.search(r'```(?:json)?\s*(.*?)```', s, flags=re.S)
+    if m:
+        s = m.group(1).strip()
+    try:
+        return json.loads(s)
+    except json.JSONDecodeError:
+        pass
+    start = s.find('{')
+    if start >= 0:
+        repaired = _repair_truncated(s[start:])
+        if repaired is not None:
+            log.warning("规划器输出疑似被截断，已自动截断修复（可能丢失末尾部分方案项）")
+            return repaired
+    raise ValueError("模型输出无法解析为 JSON")
+
+
 async def generate_plan(
     product_info: dict,
     sku_variants: list[str],
@@ -94,7 +140,7 @@ async def generate_plan(
         "目标平台": platform,            # ozon/wb/cn（决定图内文案语言）
     }, ensure_ascii=False)
 
-    async with httpx.AsyncClient(timeout=60, verify=False) as client:
+    async with httpx.AsyncClient(timeout=90, verify=False) as client:
         resp = await client.post(
             f"{s.deepseek_base_url.rstrip('/')}/chat/completions",
             headers={"Authorization": f"Bearer {s.deepseek_api_key}", "Content-Type": "application/json"},
@@ -106,16 +152,30 @@ async def generate_plan(
                 ],
                 "response_format": {"type": "json_object"},
                 "temperature": 0.3,
-                "max_tokens": 2000,
+                "max_tokens": 8000,
             },
         )
         resp.raise_for_status()
-        content = resp.json()["choices"][0]["message"]["content"]
+        body = resp.json()
+        message = body["choices"][0]["message"]
+        finish_reason = body["choices"][0].get("finish_reason", "")
+        usage = body.get("usage") or {}
+        log.info(
+            "规划器 token 用量: prompt=%s completion=%s finish=%s",
+            usage.get("prompt_tokens", "?"), usage.get("completion_tokens", "?"), finish_reason,
+        )
+        content = message.get("content") or ""
+        # 思考型输出：content 为空时从 reasoning_content 里捞
+        if not content.strip() and message.get("reasoning_content"):
+            content = message["reasoning_content"]
 
     try:
-        data = json.loads(content)
-    except json.JSONDecodeError as exc:
-        log.error("规划器输出不是合法 JSON: %s", content[:200])
+        data = _extract_json(content)
+    except ValueError as exc:
+        log.error(
+            "规划器输出解析失败 finish_reason=%s content[:200]=%s",
+            finish_reason, content[:200],
+        )
         raise RuntimeError("规划器输出解析失败") from exc
 
     items = _normalize_items(data.get("items") or [], sku_variants)

server/services/prompt.py

@@ -12,38 +12,53 @@ from __future__ import annotations
 
 import re
 
-# ── 风格模板（与插件端 STYLE_SET_OPTIONS 对应）─────────────────────────────
+# ── 风格模板（与插件端 STYLE_SET_OPTIONS 对应；提示词可被用户在插件里改写覆盖）───
+# 提示词用中文：生图 provider（通义万相/豆包）均为国产模型，中文理解一流，且便于用户自行改写。
 
 STYLE_SETS: dict[int, dict] = {
     1: {
-        "name": "经典商拍",
-        "tone": "premium commercial e-commerce photography, clean soft studio lighting, "
-                "gentle gradient background, catalog-grade presentation, refined and trustworthy",
-        "bg": "light neutral studio backdrop with soft vignette",
+        "name": "高级质感大片",
+        "tone": "高端电商大片质感，柔和的方向性棚拍光，背景带细腻的浅渐变，材质纹理清晰可见，"
+                "色彩层次高级克制，商业画册级品质，构图干净、留白充足",
+        "bg": "",
     },
     2: {
-        "name": "生活杂志",
-        "tone": "editorial lifestyle magazine aesthetic, natural window light, "
-                "cozy lived-in atmosphere, muted film tones, candid storytelling",
-        "bg": "warm lifestyle home setting with plants and textured fabrics",
+        "name": "清新生活场景",
+        "tone": "明亮通透的生活场景摄影，自然窗光，柔和的低饱和居家环境，浅景深虚化，"
+                "真实自然的氛围感，绿植与暖色织物点缀，温馨有人气",
+        "bg": "",
     },
     3: {
-        "name": "极简高冷",
-        "tone": "minimalist high-end aesthetic, vast negative space, single directional light, "
-                "cool grey palette, architectural calm, quiet luxury",
-        "bg": "seamless light grey studio background with subtle shadow",
+        "name": "极简白底规范",
+        "tone": "极简棚拍风格，纯净无缝的浅色背景，柔和均匀的无影布光，以商品为中心的严谨构图，"
+                "安静的高级感，画面只保留轻微的自然接触投影",
+        "bg": "",
     },
     4: {
-        "name": "活力爆款",
-        "tone": "vibrant high-conversion e-commerce style, punchy saturated accents, "
-                "energetic composition, bold contrast, promotional poster energy",
-        "bg": "bright colorful gradient backdrop with dynamic geometric shapes",
+        "name": "炫彩促销风",
+        "tone": "高能量促销风格，高饱和度色块背景搭配动感几何图形，强对比，节日大促海报氛围，"
+                "构图抢眼、视觉冲击力强",
+        "bg": "",
     },
     5: {
-        "name": "暗调质感",
-        "tone": "dark moody premium product photography, dramatic rim lighting, "
-                "deep charcoal background, rich texture detail, luxurious atmosphere",
-        "bg": "matte black background with soft spotlight and subtle smoke haze",
+        "name": "暗调轻奢",
+        "tone": "暗调轻奢质感，深炭灰色背景，轮廓光勾勒商品边缘，材质细节丰富，带轻微雾感，"
+                "如美术馆展陈般的呈现",
+        "bg": "",
+    },
+    6: {
+        "name": "俄式风情",
+        "tone": "俄式风情电商大片，浓郁温暖的色调，红与金的传统配色点缀，冬日节庆氛围，"
+                "深色木质与毛毡织物背景，如暖炉烛光般的柔和光晕，厚重扎实的质感，"
+                "带一丝巴洛克式的华丽细节，适合俄语区市场",
+        "bg": "",
+    },
+    7: {
+        "name": "北欧极简",
+        "tone": "北欧极简风格，白色与浅灰的原木空间，大量自然漫射光，干净利落的线条，"
+                "浅色木质背景点缀少量绿植，克制的中性配色，画面通透轻盈，"
+                "舒适宁静的氛围",
+        "bg": "",
     },
 }
 
@@ -192,9 +207,10 @@ def _prompt_material(ctx: dict, style: dict, lang: str) -> str:
     )
 
 def _prompt_lifestyle(ctx: dict, style: dict, lang: str) -> str:
+    bg = f" ({style['bg']})" if style.get("bg") else ""
     return (
         f"Lifestyle in-context scene for product \"{ctx['title']}\": the product is naturally "
-        f"used / placed in a real environment ({style['bg']}), realistic human-scale surroundings, "
+        f"used / placed in a real environment{bg}, realistic human-scale surroundings, "
         f"soft daylight, authentic candid mood, product remains the clear visual focus. "
         f"{style['tone']}. {TEXT_RENDER[lang]} {QUALITY} {PRODUCT_REF_LOCK}"
     )
@@ -242,6 +258,7 @@ def _prompt_custom(ctx: dict, style: dict, lang: str, extra: dict) -> str:
     hint = (extra.get("prompt_hint") or "").strip()
     purpose = extra.get("title") or ""
     detail = extra.get("detail") or ""
+    bg = f" {style['bg']} as environment." if style.get("bg") else ""
     composed = (
         f"E-commerce marketing image for product \"{ctx['title']}\""
         + (f" — {purpose}" if purpose else "")
@@ -250,7 +267,7 @@ def _prompt_custom(ctx: dict, style: dict, lang: str, extra: dict) -> str:
     )
     if hint:
         composed += f" Composition: {hint}."
-    return f"{composed} {style['tone']}. {style['bg']} as environment. {TEXT_RENDER[lang]} {QUALITY} {PRODUCT_REF_LOCK}"
+    return f"{composed}{bg} {style['tone']}. {TEXT_RENDER[lang]} {QUALITY} {PRODUCT_REF_LOCK}"
 
 _PROMPT_BUILDERS = {
     "white_bg": _prompt_white_bg,
@@ -265,13 +282,18 @@ _PROMPT_BUILDERS = {
 }
 
 
-def build_prompt(type_id: str, ctx: dict, style_set: int, lang: str, extra: dict | None = None) -> str:
+def build_prompt(type_id: str, ctx: dict, style_set: int, lang: str, extra: dict | None = None,
+                 style_prompt: str | None = None) -> str:
     """构造指定图类型的完整生图 prompt。
 
     extra: 方案项信息 {title, detail, prompt_hint}——custom 类型必需，
     预设类型也会把 prompt_hint 作为构图补充注入。
+    style_prompt: 用户改写的风格提示词，覆盖 style_set 内置模板（tone/bg 整体替换）。
     """
-    style = STYLE_SETS.get(style_set, STYLE_SETS[1])
+    if style_prompt and style_prompt.strip():
+        style = {"name": "custom", "tone": style_prompt.strip(), "bg": ""}
+    else:
+        style = STYLE_SETS.get(style_set, STYLE_SETS[1])
     extra = extra or {}
     if type_id == "custom":
         prompt = _prompt_custom(ctx, style, lang, extra)