feat: 调用gpt\nano模型方式修改，功能优化，增加下载采集图片等功能

2026-08-20 17:00:40 +08:00
parent 90b7c8737d
commit 95cb93a160
20 changed files with 411 additions and 407 deletions
@@ -0,0 +1,120 @@
+"""导出采集图片：把采集到的源站图片打包成 ZIP 下载到本地。
+
+参考图 URL 可能是源站 CDN（需 Referer 绕过防盗链）或本地上传的 media 文件。
+ZIP 内部结构沿用现有分组名建子文件夹（主图 / SKU图片 / 详情图 / 手动上传），
+文件名沿用采集 key（main-001 等）+ SKU 规格名；顶层文件夹用商品标题（清洗后）。
+"""
+from __future__ import annotations
+
+import io
+import mimetypes
+import re
+import zipfile
+
+from fastapi import APIRouter, HTTPException
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel, Field
+
+from api.proxy import guess_referer
+from services import storage
+
+router = APIRouter(prefix="/api", tags=["export"])
+
+_EXT_BY_CTYPE = {
+    "image/jpeg": ".jpg",
+    "image/png": ".png",
+    "image/webp": ".webp",
+    "image/gif": ".gif",
+    "image/bmp": ".bmp",
+}
+
+
+class ExportImageItem(BaseModel):
+    url: str
+    groupName: str = "主图"
+    variantName: str | None = None
+    key: str = ""  # 采集 key，如 main-001 / sku-002 / upload-001
+
+
+class ExportImagesRequest(BaseModel):
+    title: str | None = Field(default=None, description="商品标题，用作 ZIP 顶层文件夹名")
+    images: list[ExportImageItem]
+
+
+def _clean(name: str) -> str:
+    """清洗文件夹/文件名非法字符（与插件 cleanFilename 同规则，Windows 兼容）。"""
+    s = re.sub(r'[<>:"/\\|?*\x00-\x1f]', "_", (name or "").strip())
+    s = re.sub(r"\s+", "_", s)
+    return s.strip(" .")[:80]
+
+
+def _ext(url: str, ctype: str) -> str:
+    """由 content-type（优先）或 URL 后缀决定扩展名。"""
+    ctype = ctype.split(";")[0].strip().lower()
+    if ctype in _EXT_BY_CTYPE:
+        return _EXT_BY_CTYPE[ctype]
+    if ctype.startswith("image/"):
+        return "." + ctype.split("/")[-1]
+    path = url.split("?")[0].lower()
+    for ext in (".jpg", ".jpeg", ".png", ".webp", ".gif", ".bmp"):
+        if path.endswith(ext):
+            return ".jpg" if ext == ".jpeg" else ext
+    return ".jpg"
+
+
+def _is_image(url: str, ctype: str) -> bool:
+    ctype = ctype.split(";")[0].strip().lower()
+    if ctype.startswith("image/"):
+        return True
+    return bool(re.search(r"\.(jpe?g|png|webp|gif|bmp)(\?|$)", url, re.IGNORECASE))
+
+
+async def _download(url: str) -> tuple[bytes, str]:
+    """本地 media 文件直读磁盘；远程 URL 带 Referer 下载。"""
+    path = storage.local_path(url)
+    if path is not None:
+        mime = mimetypes.guess_type(path.name)[0] or "image/jpeg"
+        return path.read_bytes(), mime
+    return await storage.download_bytes(url, referer=guess_referer(url))
+
+
+@router.post("/export-images")
+async def export_images(req: ExportImagesRequest):
+    if not req.images:
+        raise HTTPException(status_code=400, detail="没有可导出的图片")
+
+    root = _clean(req.title) or "采集图片"
+    buf = io.BytesIO()
+    used: set[str] = set()
+
+    with zipfile.ZipFile(buf, "w", zipfile.ZIP_DEFLATED) as zf:
+        for img in req.images:
+            try:
+                data, ctype = await _download(img.url)
+            except Exception:  # noqa: BLE001
+                continue  # 单张失败不中断整包
+            if not _is_image(img.url, ctype):
+                continue
+
+            ext = _ext(img.url, ctype)
+            base = _clean(img.key) or "image"
+            if img.variantName:
+                base += f"-{_clean(img.variantName)}"
+            filename = f"{base}{ext}"
+            if filename in used:  # 同名加序号防覆盖
+                stem = filename[: -len(ext)]
+                n = 2
+                while f"{stem}-{n}{ext}" in used:
+                    n += 1
+                filename = f"{stem}-{n}{ext}"
+            used.add(filename)
+
+            group = _clean(img.groupName) or "图片"
+            zf.writestr(f"{root}/{group}/{filename}", data)
+
+    buf.seek(0)
+    return StreamingResponse(
+        buf,
+        media_type="application/zip",
+        headers={"Content-Disposition": 'attachment; filename="collect.zip"'},
+    )
@@ -41,14 +41,12 @@ class Settings(BaseSettings):
    dashscope_base_url: str = ""        # 留空按模型自动选择万象异步/千问同步端点
    dashscope_model: str = "wan2.7-image-pro"

-    # RightAPI（OpenAI 兼容中转，gpt-image 系列）
+    # RightAPI（OpenAI 兼容中转，gpt-image / nano-banana 系列）
    rightapi_api_key: str = ""
    rightapi_base_url: str = "https://rightapi.ai/draw"
    rightapi_image_model: str = "gpt-image-2"
-    rightapi_image_quality: str = "high"  # auto | low | medium | high
    rightapi_max_retries: int = 3         # 429/5xx/超时的重试次数（1 = 不重试）
    rightapi_retry_wait: int = 60         # 重试基础等待秒数，按 60→120→240 递增
-    rightapi_input_fidelity: str = "high"  # edits 端点高保真档（high | low，留空关闭）

    # DeepSeek（出图方案规划器）
    deepseek_api_key: str = ""
@@ -7,7 +7,7 @@ from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles

-from api import generate, proxy, suites, upload
+from api import export, generate, proxy, suites, upload
 from config import get_settings
 from services.storage import media_root

@@ -26,6 +26,7 @@ app.include_router(generate.router)
 app.include_router(suites.router)
 app.include_router(proxy.router)
 app.include_router(upload.router)
+app.include_router(export.router)

 # 静态托管生成的图片
 app.mount("/media", StaticFiles(directory=str(media_root())), name="media")
@@ -234,71 +234,103 @@ async def generate_tongyi(prompt: str, ref_images: list[str], size: str = "2048*
        return dl.content


-# ── Provider：RightAPI（gpt-image，OpenAI 兼容中转）──────────────────────
+# ── Provider：RightAPI（gpt-image / nano-banana，OpenAI 兼容中转）──────────

 # 可重试的状态码：中转限流/网关抖动（该中转限流时返回 Cloudflare 502 而非 429）
 RETRYABLE_STATUS = {429, 500, 502, 503, 504}

-# 中转对 input_fidelity 参数的支持探测：按模型记忆不支持该参数的模型（gpt-image 系列支持，
-# nano-banana 系列可能不认；降级只影响触发过的模型，不牵连其他模型）
-_rightapi_fidelity_unsupported: set[str] = set()
+
+async def _rightapi_poll_task(client: httpx.AsyncClient, headers: dict, origin: str,
+                              task_id: str, max_wait: int) -> dict:
+    """轮询站点级任务查询接口 GET /v1/tasks/{task_id}（不带 /draw 前缀）。
+
+    实测要点（docs/rightapi-调用排查与修复方案.md §2.2）：
+      - 完成响应**没有** status:"completed" 字段，完成判定 = 响应里出现 data；
+      - progress 基本不动（0~2），不能当进度条依据；
+      - 失败态 = status 为 failed / error / cancelled。
+    """
+    poll_url = f"{origin}/v1/tasks/{task_id}"
+    elapsed, interval = 0, 3
+    while elapsed < max_wait:
+        resp = await client.get(poll_url, headers=headers, timeout=30)
+        _raise_api_error(resp, "RightAPI")
+        result = resp.json()
+        status = result.get("status", "")
+        if status in ("failed", "error", "cancelled"):
+            err = result.get("error") or {}
+            raise RuntimeError(f"RightAPI 任务失败: {err.get('message') or result}")
+        if "data" in result:
+            return result
+        await asyncio.sleep(interval)
+        elapsed += interval
+        interval = min(interval + 2, 10)
+    raise TimeoutError(f"RightAPI 异步任务超时 ({max_wait}s): task_id={task_id}")
+
+
+def _rightapi_extract_image(result: dict) -> tuple[str | None, str | None]:
+    """从轮询完成结果里取 (kind, payload)：kind ∈ url | b64，未取到返回 (None, None)。
+
+    完成形状为 Images 协议：{"created":..., "data":[{"url": "..."}]}（实测只见 url）。
+    """
+    data = result.get("data") or []
+    if data:
+        item = data[0] or {}
+        url = item.get("url") or ""
+        if url:
+            return ("url", url)
+        b64 = item.get("b64_json") or ""
+        if b64:
+            return ("b64", b64)
+    return (None, None)


 async def _rightapi_request(s, prompt: str, ref_images: list[str], size: str, model: str) -> bytes:
-    """RightAPI 各模型：有参考图走 /v1/images/edits（multipart），无参考图走 /v1/images/generations。
+    """RightAPI 各模型：统一走 /v1/images/generations（异步）。

-    OpenAI 兼容协议：响应固定 b64_json（不支持 response_format 参数，传了报 400）；
-    同步调用无任务轮询，高质量档单张 1-5 分钟，超时按文档建议兜底 600s。
-    input_fidelity=high 是 gpt-image-1 的 edits 保真参数（gpt-image-2 官方已移除、默认高保真，
-    官逆通道更是不识别）；带上是为了兼容按 gpt-image-1 语义实现的中转，中转不认（400）则按模型
-    自动去掉重试并记住，该模型后续请求不再带。
+    官方协议（docs.rightapi.ai/docs/rc_draw/，2026-07 起统一异步）：
+      - POST /draw/v1/images/generations，请求体固定带 async:true，参考图放 image 数组（data-URI）；
+      - 返回 task_id 后轮询 GET /v1/tasks/{task_id}（站点级，不带 /draw）；
+      - 参数只有 model/prompt/n/size/imageSize/image/async；不传 quality/output_format/input_fidelity。
+    参考图沿用现选图逻辑（≤2 张，image 数组）。单张 1-5 分钟，轮询上限 poll_max_wait 兜底。
    """
    base = s.rightapi_base_url.rstrip("/")
+    # 任务查询是站点级接口，不带 /draw：从 base 里拆出 origin（https://rightapi.ai/draw → https://rightapi.ai）
+    origin = base.split("/draw", 1)[0].rstrip("/") or base
    headers = {"Authorization": f"Bearer {s.rightapi_api_key}"}
-    use_fidelity = (
-        bool(ref_images) and s.rightapi_input_fidelity and model not in _rightapi_fidelity_unsupported
-    )
+
+    body = {
+        "model": model,
+        "prompt": prompt,
+        "n": 1,
+        "size": size,
+        "async": True,
+    }
+    if ref_images:
+        body["image"] = [await _resolve_ref(u) for u in ref_images]

    async with httpx.AsyncClient(timeout=max(s.request_timeout, 600), verify=False) as client:
-        common = {
-            "model": model,
-            "prompt": prompt,
-            "size": size,
-            "quality": s.rightapi_image_quality,
-            "output_format": "jpeg",  # 与落盘 .jpg 后缀一致
-            "n": 1,
-        }
-        if use_fidelity:
-            common["input_fidelity"] = s.rightapi_input_fidelity
-        if ref_images:
-            files = []
-            for i, u in enumerate(ref_images):
-                data, mime = await _resolve_ref_bytes(u)
-                files.append(("image[]", (f"ref-{i + 1}.{mime.split('/')[-1]}", data, mime)))
-            resp = await client.post(f"{base}/v1/images/edits", headers=headers, files=files, data=common)
-            # 中转不认 input_fidelity：去掉参数重试一次（仅一次探测），降级只记到当前模型
-            if resp.status_code == 400 and use_fidelity and "input_fidelity" in resp.text:
-                _rightapi_fidelity_unsupported.add(model)
-                log.warning("RightAPI 模型 %s 不支持 input_fidelity 参数，已自动去掉并降级（该模型后续请求不再带）", model)
-                common.pop("input_fidelity", None)
-                resp = await client.post(f"{base}/v1/images/edits", headers=headers, files=files, data=common)
-        else:
-            resp = await client.post(
-                f"{base}/v1/images/generations",
-                headers={**headers, "Content-Type": "application/json"},
-                json=common,
-            )
+        resp = await client.post(
+            f"{base}/v1/images/generations",
+            headers={**headers, "Content-Type": "application/json"},
+            json=body,
+        )
        _raise_api_error(resp, "RightAPI")
-        item = resp.json()["data"][0]
-        b64 = item.get("b64_json") or ""
-        if b64:
-            return base64.b64decode(b64.split(",", 1)[-1] if "," in b64 else b64)
-        img_url = item.get("url") or ""
-        if not img_url:
-            raise RuntimeError(f"RightAPI 响应里没有图片数据: {item}")
-        dl = await client.get(img_url, timeout=s.request_timeout)
-        dl.raise_for_status()
-        return dl.content
+        submitted = resp.json()
+        task_id = submitted.get("task_id") or ""
+        if task_id:
+            result = await _rightapi_poll_task(client, headers, origin, task_id, s.poll_max_wait)
+        else:
+            # 极端兜底：个别中转可能同步返回 data（文档不保证，但防御处理）
+            result = submitted
+
+        kind, payload = _rightapi_extract_image(result)
+        if kind == "b64" and payload:
+            return base64.b64decode(payload.split(",", 1)[-1] if "," in payload else payload)
+        if kind == "url" and payload:
+            dl = await client.get(payload, timeout=s.request_timeout)
+            dl.raise_for_status()
+            return dl.content
+        raise RuntimeError(f"RightAPI 任务完成但没有图片数据: {result}")


 async def generate_rightapi(prompt: str, ref_images: list[str], size: str = "2048x2048", model: str | None = None) -> bytes: