feat(product-suite): add cmhub vision AI writing

This commit is contained in:
chengma
2026-07-17 09:09:09 +08:00
parent 56d6b59a00
commit 40a9f5fbf2
15 changed files with 575 additions and 40 deletions
+127 -6
View File
@@ -47,6 +47,10 @@ class CMHubError(AIError):
CMHUB_IMAGE_MAX_BYTES = 20 * 1024 * 1024
CMHUB_TITLE_READ_TIMEOUT_SECONDS = 600
CMHUB_VISION_READ_TIMEOUT_SECONDS = 120
CMHUB_VISION_MAX_IMAGES = 8
CMHUB_VISION_MAX_IMAGE_BYTES = 10 * 1024 * 1024
CMHUB_VISION_MAX_TOTAL_BYTES = 32 * 1024 * 1024
CMHUB_IMAGE_READ_TIMEOUT_SECONDS = 900
CMHUB_IMAGE_SUBMIT_READ_TIMEOUT_SECONDS = 36
CMHUB_IMAGE_POLL_READ_TIMEOUT_SECONDS = 15
@@ -162,6 +166,55 @@ def gen_title(
return text
def analyze_product_images(
instruction,
context,
image_paths,
*,
config=None,
cmhub_config_path=appconfig.CMHUB_CONFIG_PATH,
on_event=None,
):
"""Use the dedicated cmhub vision route for product-suite AI writing."""
cfg = appconfig.load_config() if config is None else config
ai_cfg = appconfig.ai_config(cfg)
if _ai_backend(ai_cfg) != "cmhub":
raise AIError("商品套图AI帮写需要使用 cmhub 图片理解服务,请到⑤设置配置 cmhub。")
paths = _validate_cmhub_vision_images(image_paths)
runtime = _cmhub_runtime(cfg, "vision", cmhub_config_path)
payload = {
"prompt": _compose_product_suite_vision_prompt(instruction, context),
"model": runtime["alias"],
"images": [{"image_base64": _image_data_url(path)} for path in paths],
"parameters": {"temperature": 0.2},
}
try:
data = _cmhub_call_with_retry(
"POST",
appconfig.cmhub_request_url(runtime["base_url"], "/api/v1/analyze/images"),
runtime["api_key"],
payload=payload,
connect_timeout=runtime["connect_timeout"],
read_timeout=CMHUB_VISION_READ_TIMEOUT_SECONDS,
attempts=1,
on_retry=None,
)
except CMHubError as exc:
raise _vision_cmhub_error(exc) from exc
_emit_cmhub_metadata(on_event, data, "vision_request")
text = _extract_text(data).strip()
if not text:
raise AIError("图片理解服务未返回可用卖点,请稍后重试")
return {
"text": text,
"image_count": len(paths),
"metadata": _cmhub_metadata(data),
}
def gen_cover(
cover_prompt,
old_cover_path,
@@ -1338,17 +1391,67 @@ def _download_and_save_cmhub_cover(request_result, on_step=None):
return saved_path
def _validate_cmhub_vision_images(image_paths):
paths = [str(path or "").strip() for path in list(image_paths or [])]
if not paths:
raise AIError("请先添加至少一张可用商品原图")
if len(paths) > CMHUB_VISION_MAX_IMAGES:
raise AIError("图片理解最多支持%d张商品原图" % CMHUB_VISION_MAX_IMAGES)
total_size = 0
for index, path in enumerate(paths, 1):
if not path or not os.path.isfile(path):
raise AIError("第%d张商品原图尚未下载完成,请稍后重试" % index)
try:
size = os.path.getsize(path)
except OSError as exc:
raise AIError("第%d张商品原图无法读取,请稍后重试" % index) from exc
if size > CMHUB_VISION_MAX_IMAGE_BYTES:
raise AIError("第%d张商品原图超过10MiB,无法进行AI帮写" % index)
total_size += size
if total_size > CMHUB_VISION_MAX_TOTAL_BYTES:
raise AIError("商品原图总大小超过32MiB,无法进行AI帮写")
return paths
def _vision_cmhub_error(exc):
code = str(getattr(exc, "code", "") or "unknown")
if code == "read_timeout":
message = "等待图片理解结果超时,结果未确认,请先查看点数余额或稍后重试"
elif code in {"connect_timeout", "network_error"}:
message = "连接图片理解服务失败,结果未确认,请检查网络后稍后重试"
else:
message = _cmhub_user_message(code, "")
if not message or message == code:
message = "图片理解服务暂时不可用,请稍后重试"
return CMHubError(
code,
message,
status=getattr(exc, "status", None),
retryable=False,
retry_after=getattr(exc, "retry_after", None),
)
def _cmhub_runtime(config, operation, cmhub_config_path):
hub = appconfig.cmhub_config(config)
api_key = appconfig.get_cmhub_api_key(path=cmhub_config_path)
alias_key = "title_alias" if operation == "title" else "image_alias"
operation_config = {
"title": ("title_alias", "生文别名"),
"image": ("image_alias", "生图别名"),
"vision": ("vision_alias", "图片理解别名"),
}
try:
alias_key, alias_label = operation_config[str(operation or "")]
except KeyError as exc:
raise AIError("cmhub 操作类型无效") from exc
missing = []
if not hub.get("base_url"):
missing.append("Base URL")
if not api_key:
missing.append("API Key")
if not hub.get(alias_key):
missing.append("生文别名" if operation == "title" else "生图别名")
missing.append(alias_label)
if missing:
raise CMHubError(
"cmhub_not_configured",
@@ -1379,6 +1482,18 @@ def _compose_title_prompt(title_prompt, old_title):
return "请只返回新标题,不要解释。"
def _compose_product_suite_vision_prompt(instruction, context):
return (
"请分析当前电商商品原图,并根据图片可见信息生成可直接编辑的「商品卖点与要求」。"
"请使用任务指定的输出语言,只输出卖点与画面要求正文,不要解释分析过程。"
"内容应包含商品名称或品类、颜色、款式、可见细节、核心卖点、目标人群、使用场景"
"和适合套图生成的画面要求。已有要求仅是补充约束,不能当作图片事实。"
"不要虚构材质、尺寸、功能、认证、价格、物流承诺或图片中不可确认的信息。"
"\n\n任务要求:\n%s\n\n任务上下文:\n%s"
% (str(instruction or "").strip(), str(context or "").strip())
)
def _normalize_cmhub_resolution(resolution):
value = str(resolution or "1k").strip().lower()
mapping = {
@@ -1697,14 +1812,20 @@ def _notify_cmhub_retry(callback, step, attempt, attempts, exc):
pass
def _emit_cmhub_metadata(callback, data, step):
if callback is None or not isinstance(data, dict):
return
metadata = {
def _cmhub_metadata(data):
if not isinstance(data, dict):
return {}
return {
key: data.get(key)
for key in ("alias", "model_used", "points_cost", "points_balance", "call_id")
if data.get(key) is not None
}
def _emit_cmhub_metadata(callback, data, step):
if callback is None:
return
metadata = _cmhub_metadata(data)
if not metadata:
return
try: