Skip to content

利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.

fix(material): bound generated image downloads and decoding

Rudy Celeklif7cf72e1641a
diff --git a/app/services/material.py b/app/services/material.py
index eb98f49b..807d55b0 100644
--- a/app/services/material.py
+++ b/app/services/material.py
@@ -7,6 +7,7 @@ import tempfile
import threading
import time
import uuid
+import warnings
from pathlib import Path
from typing import Any, Callable, List
from urllib.parse import quote_plus, urlencode, urlsplit, urlunsplit
@@ -1209,6 +1210,8 @@ OPENAI_IMAGE_RETRYABLE_STATUS_CODES = frozenset({429, 500, 502, 503, 504})
# key 时重试会自动换 key;只有一个 key 时快速失败,不做无意义重试。
OPENAI_IMAGE_KEY_ERROR_STATUS_CODES = frozenset({401, 403})
OPENAI_IMAGE_MAX_ATTEMPTS = 3
+OPENAI_IMAGE_MAX_BYTES = 25 * 1024 * 1024
+OPENAI_IMAGE_MAX_PIXELS = 50_000_000
# 串行出图 + 线性退避,兼容中转服务普遍的限流恢复窗口。
OPENAI_IMAGE_RETRY_BACKOFF_SECONDS = (5, 15, 30)
# 同步生成接口可能需要数十秒才返回图片,读超时给足余量。
@@ -1338,6 +1341,7 @@ def _openai_image_download_bytes(
"""
failure_detail = "no download attempt was made"
for attempt in range(1, OPENAI_IMAGE_MAX_DOWNLOAD_ATTEMPTS + 1):
+ response = None
try:
response = requests.get(
image_url,
@@ -1349,14 +1353,40 @@ def _openai_image_download_bytes(
proxies=config.proxy,
verify=_get_tls_verify(),
timeout=(30, 120),
+ stream=True,
)
- if response.status_code == 200 and response.content:
- return response.content, ""
- failure_detail = f"HTTP {response.status_code} while downloading image"
+ if response.status_code == 200:
+ try:
+ declared_size = int(response.headers.get("Content-Length", 0))
+ except (TypeError, ValueError):
+ declared_size = 0
+ if declared_size > OPENAI_IMAGE_MAX_BYTES:
+ return None, "generated image exceeds the 25 MB download limit"
+ image_bytes = bytearray()
+ for chunk in response.iter_content(chunk_size=64 * 1024):
+ if not chunk:
+ continue
+ if len(image_bytes) + len(chunk) > OPENAI_IMAGE_MAX_BYTES:
+ return None, "generated image exceeds the 25 MB download limit"
+ image_bytes.extend(chunk)
+ if image_bytes:
+ return bytes(image_bytes), ""
+ failure_detail = "generated image download was empty"
+ else:
+ failure_detail = f"HTTP {response.status_code} while downloading image"
except Exception as e:
failure_detail = (
f"error={type(e).__name__}, detail={_redact_request_error(e, api_key)}"
)
+ finally:
+ if response is not None:
+ try:
+ response.close()
+ except Exception as exc:
+ logger.warning(
+ "failed to close generated image download response: "
+ f"error={type(exc).__name__}"
+ )
if attempt < OPENAI_IMAGE_MAX_DOWNLOAD_ATTEMPTS:
logger.warning(
"generated image download failed, retrying the same url: "
@@ -1388,10 +1418,18 @@ def _parse_openai_image_response(
b64_payload = entry.get("b64_json")
if b64_payload:
+ if (
+ not isinstance(b64_payload, str)
+ or len(b64_payload) > 4 * ((OPENAI_IMAGE_MAX_BYTES + 2) // 3)
+ ):
+ return None, "generated image exceeds the 25 MB response limit"
try:
- return base64.b64decode(b64_payload), ""
+ image_bytes = base64.b64decode(b64_payload)
except Exception as e:
return None, f"invalid b64_json payload: {type(e).__name__}"
+ if len(image_bytes) > OPENAI_IMAGE_MAX_BYTES:
+ return None, "generated image exceeds the 25 MB response limit"
+ return image_bytes, ""
image_url = entry.get("url")
if isinstance(image_url, str) and image_url.startswith(("http://", "https://")):
@@ -1506,21 +1544,34 @@ def _save_openai_image_file(
os.makedirs(save_dir, exist_ok=True)
image_path = os.path.join(save_dir, f"openai-image-{uuid.uuid4().hex[:12]}.png")
+ if len(image_bytes) > OPENAI_IMAGE_MAX_BYTES:
+ raise _OpenAIImageDecodeError("generated image exceeds the 25 MB limit")
# 图片解码失败可以降级为“跳过当前关键词”,但目录权限、磁盘空间和文件
# 写入失败必须继续抛出,否则按需生成循环会在本地无法保存文件时继续创建
# 后续付费任务。Image.open 只读取内存字节,因此这里的 OSError 属于格式
# 识别失败;image.load 的 OSError 则对应截断或损坏的图片数据。
+ image = None
try:
- image = Image.open(io.BytesIO(image_bytes))
- except (UnidentifiedImageError, OSError, SyntaxError, ValueError) as exc:
+ with warnings.catch_warnings():
+ warnings.simplefilter("error", Image.DecompressionBombWarning)
+ image = Image.open(io.BytesIO(image_bytes))
+ if image.width * image.height > OPENAI_IMAGE_MAX_PIXELS:
+ raise ValueError("generated image exceeds the 50 million pixel limit")
+ image.load()
+ except (
+ Image.DecompressionBombWarning,
+ Image.DecompressionBombError,
+ UnidentifiedImageError,
+ OSError,
+ SyntaxError,
+ ValueError,
+ ) as exc:
+ if image is not None:
+ image.close()
raise _OpenAIImageDecodeError(f"{type(exc).__name__}: {exc}") from exc
with image:
- try:
- image.load()
- except (OSError, SyntaxError, ValueError) as exc:
- raise _OpenAIImageDecodeError(f"{type(exc).__name__}: {exc}") from exc
if image.mode not in ("RGB", "RGBA", "L", "LA", "P"):
image = image.convert("RGB")
# save 不放进解码异常保护区:写入错误表示运行环境持续不可用,应立即
diff --git a/test/services/test_material_openai_image.py b/test/services/test_material_openai_image.py
index 15c6f43f..6ae33a77 100644
--- a/test/services/test_material_openai_image.py
+++ b/test/services/test_material_openai_image.py
@@ -5,6 +5,7 @@ import os
import shutil
import tempfile
import unittest
+import warnings
from types import SimpleNamespace
from unittest.mock import patch
@@ -26,7 +27,13 @@ def _image_response(payload, status_code=200):
def _download_response(content, status_code=200):
- return SimpleNamespace(status_code=status_code, content=content)
+ return SimpleNamespace(
+ status_code=status_code,
+ content=content,
+ headers={"Content-Length": str(len(content))},
+ iter_content=lambda chunk_size: iter((content,)),
+ close=lambda: None,
+ )
class TestOpenAIImageProvider(unittest.TestCase):
@@ -378,6 +385,66 @@ class TestOpenAIImageProvider(unittest.TestCase):
for call in get.call_args_list:
self.assertEqual(call.args[0], "https://cdn.example.com/generated/x.png")
+ def test_generated_image_download_rejects_oversize_body(self):
+ """A generated image URL must not buffer an unbounded provider body."""
+ response = _download_response(b"too-large")
+ response.headers = {}
+ closed = []
+ response.close = lambda: closed.append(True)
+ with (
+ patch("app.services.material.requests.get", return_value=response) as get,
+ patch("app.services.material.OPENAI_IMAGE_MAX_BYTES", 5, create=True),
+ patch("app.services.material.time.sleep"),
+ ):
+ image_bytes, error = material._openai_image_download_bytes(
+ "https://cdn.example.com/generated/x.png", "test-key"
+ )
+
+ self.assertIsNone(image_bytes)
+ self.assertIn("limit", error)
+ self.assertTrue(all(call.kwargs.get("stream") is True for call in get.call_args_list))
+ self.assertEqual(closed, [True])
+
+ def test_generated_image_download_rejects_declared_oversize_body(self):
+ """An advertised oversize image should be rejected before reading bytes."""
+ response = _download_response(b"small")
+ response.headers = {"Content-Length": "100"}
+ response.iter_content = lambda chunk_size: self.fail("body should not be read")
+ with (
+ patch("app.services.material.requests.get", return_value=response),
+ patch("app.services.material.OPENAI_IMAGE_MAX_BYTES", 5),
+ ):
+ image_bytes, error = material._openai_image_download_bytes(
+ "https://cdn.example.com/generated/x.png", "test-key"
+ )
+
+ self.assertIsNone(image_bytes)
+ self.assertIn("limit", error)
+
+ def test_generated_image_response_rejects_oversize_base64(self):
+ """Do not decode an unbounded inline image from a provider response."""
+ response = _image_response(
+ {"data": [{"b64_json": base64.b64encode(b"too-large").decode()}]}
+ )
+ with patch("app.services.material.OPENAI_IMAGE_MAX_BYTES", 5):
+ image_bytes, error = material._parse_openai_image_response(response, "test-key")
+
+ self.assertIsNone(image_bytes)
+ self.assertIn("limit", error)
+
+ def test_generated_image_decode_rejects_pillow_bomb_warning(self):
+ """Small compressed bytes can still expand to too many image pixels."""
+ image_bytes = _png_bytes(width=12, height=12)
+ with (
+ patch.object(Image, "MAX_IMAGE_PIXELS", 100),
+ warnings.catch_warnings(),
+ ):
+ warnings.simplefilter("ignore", Image.DecompressionBombWarning)
+ with self.assertRaises(material._OpenAIImageDecodeError):
+ material._save_openai_image_file(image_bytes, self.save_dir)
+
+ self.assertEqual(os.listdir(self.save_dir), [])
+
def test_generate_images_openai_returns_empty_on_rejected_request(self):
"""业务拒绝(如内容策略)返回空结果,不做退避重试。"""
response = _image_response(