GPT image 2 无法编辑图片，只能生成
- 内容介绍
- 文章标签
- 相关推荐
我发现不论是 cliproxyapi还是 sub2api 这俩平台转换出来的API，不论是response协议，还是image API协议，都不能正常的编辑图片。
报错如下：
{"error":{"message":"stream disconnected before completion","type":"server_error","code":"internal_server_error"}}
但是生图都挺正常的，不知道是什么原因导致的，挺无语。
更新一下，通过调试CLI Proxy API，我得到了详细的报错。
报错的理由是OpenAI那边检测不通过，理由是色情。我挺诧异的，其实就是一张美女的图片，而且这张图片是ChatGPT自己生成的，现在编辑的时候它又说色情，真是无语了。我换了一张普通的动物照片，同样的提示词就成功了。
网友解答：--【壹】--：
文生图和图生图的接口地址好像不同，佬编辑用的是：/v1/images/edits 这个地址吗
--【贰】--：
好的，谢谢佬的建议。 审核确实是个谜，哈哈。
--【叁】--：
对，我用的是这个接口的。我现在调试 cliproxyapi，得到一个真正的来自OpenAI的报错。
[2026-04-28 20:11:12] [--------] [info ] [openai_images_handlers.go:628] openai images edit multipart parsed model=gpt-image-2 image_content_types=[image/png] image_count=1 image_filenames=[role.png] image_total_bytes=1486816 mask_count=0 method=POST output_format=png path=/v1/images/edits remote_ip=127.0.0.1 response_format=b64_json size=1024x1024 stream=false
[2026-04-28 20:11:12] [--------] [info ] [openai_images_handlers.go:652] openai images responses upstream request started model=gpt-5.4 method=POST path=/v1/images/edits remote_ip=127.0.0.1 request_bytes=1982940 response_format=b64_json stream=false
[2026-04-28 20:11:12] [2b7f8430] [debug] [conductor.go:3589] Use OAuth provider=codex auth_file=codex-k01092953640@gmail.com-plus.json for model gpt-5.4
[2026-04-28 20:11:12] [--------] [debug] [apply.go:155] thinking: original config from request | provider=codex model=gpt-5.4 mode=level budget=0 level=medium
[2026-04-28 20:11:12] [--------] [debug] [apply.go:196] thinking: processed config to apply | provider=codex model=gpt-5.4 mode=level budget=0 level=medium
[2026-04-28 20:11:15] [--------] [info ] [antigravity_version.go:76] fetched latest antigravity version version=1.23.2
[2026-04-28 20:12:22] [--------] [error] [openai_images_handlers.go:681] openai images non-streaming request failed model=gpt-5.4 error=moderation_blocked: Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual]. duration_ms=70607 method=POST path=/v1/images/edits remote_ip=127.0.0.1 response_format=b64_json status_code=502 upstream_bytes=9169 upstream_chunks=22 upstream_completed=false upstream_events=11 upstream_first_chunk_at=2026-04-28T20:11:14.747069+08:00 upstream_frames=11 upstream_last_chunk_at=2026-04-28T20:12:22.778782+08:00 upstream_last_error_code=moderation_blocked upstream_last_error_message=Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual]. upstream_last_event=response.failed
[2026-04-28 20:12:22] [2b7f8430] [error] [gin_logger.go:90] 502 | 1m10s | 127.0.0.1 | POST "/v1/images/edits"
error=moderation_blocked: Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual].
可能是我提示词有问题，但是不至于色情吧，我换别的试试。
这个详细报错，原版cliproxyapi是没有的，它只会等到超时才统一 502错误兜底。
--【肆】--：
佬，cpa要怎么打出详细的错误log的。
--【伍】--： andy6688:
moderation
moderation参数传low试试，审核级别默认是auto，传low估计好一些。不过我一直传low的都有概率遇到你那个问题。gpt的审核一直都这么迷
我发现不论是 cliproxyapi还是 sub2api 这俩平台转换出来的API，不论是response协议，还是image API协议，都不能正常的编辑图片。
报错如下：
{"error":{"message":"stream disconnected before completion","type":"server_error","code":"internal_server_error"}}
但是生图都挺正常的，不知道是什么原因导致的，挺无语。
更新一下，通过调试CLI Proxy API，我得到了详细的报错。
报错的理由是OpenAI那边检测不通过，理由是色情。我挺诧异的，其实就是一张美女的图片，而且这张图片是ChatGPT自己生成的，现在编辑的时候它又说色情，真是无语了。我换了一张普通的动物照片，同样的提示词就成功了。
网友解答：--【壹】--：
文生图和图生图的接口地址好像不同，佬编辑用的是：/v1/images/edits 这个地址吗
--【贰】--：
好的，谢谢佬的建议。 审核确实是个谜，哈哈。
--【叁】--：
对，我用的是这个接口的。我现在调试 cliproxyapi，得到一个真正的来自OpenAI的报错。
[2026-04-28 20:11:12] [--------] [info ] [openai_images_handlers.go:628] openai images edit multipart parsed model=gpt-image-2 image_content_types=[image/png] image_count=1 image_filenames=[role.png] image_total_bytes=1486816 mask_count=0 method=POST output_format=png path=/v1/images/edits remote_ip=127.0.0.1 response_format=b64_json size=1024x1024 stream=false
[2026-04-28 20:11:12] [--------] [info ] [openai_images_handlers.go:652] openai images responses upstream request started model=gpt-5.4 method=POST path=/v1/images/edits remote_ip=127.0.0.1 request_bytes=1982940 response_format=b64_json stream=false
[2026-04-28 20:11:12] [2b7f8430] [debug] [conductor.go:3589] Use OAuth provider=codex auth_file=codex-k01092953640@gmail.com-plus.json for model gpt-5.4
[2026-04-28 20:11:12] [--------] [debug] [apply.go:155] thinking: original config from request | provider=codex model=gpt-5.4 mode=level budget=0 level=medium
[2026-04-28 20:11:12] [--------] [debug] [apply.go:196] thinking: processed config to apply | provider=codex model=gpt-5.4 mode=level budget=0 level=medium
[2026-04-28 20:11:15] [--------] [info ] [antigravity_version.go:76] fetched latest antigravity version version=1.23.2
[2026-04-28 20:12:22] [--------] [error] [openai_images_handlers.go:681] openai images non-streaming request failed model=gpt-5.4 error=moderation_blocked: Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual]. duration_ms=70607 method=POST path=/v1/images/edits remote_ip=127.0.0.1 response_format=b64_json status_code=502 upstream_bytes=9169 upstream_chunks=22 upstream_completed=false upstream_events=11 upstream_first_chunk_at=2026-04-28T20:11:14.747069+08:00 upstream_frames=11 upstream_last_chunk_at=2026-04-28T20:12:22.778782+08:00 upstream_last_error_code=moderation_blocked upstream_last_error_message=Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual]. upstream_last_event=response.failed
[2026-04-28 20:12:22] [2b7f8430] [error] [gin_logger.go:90] 502 | 1m10s | 127.0.0.1 | POST "/v1/images/edits"
error=moderation_blocked: Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 4016cdac-279d-4717-935d-a2d376d12a25. safety_violations=[sexual].
可能是我提示词有问题，但是不至于色情吧，我换别的试试。
这个详细报错，原版cliproxyapi是没有的，它只会等到超时才统一 502错误兜底。
--【肆】--：
佬，cpa要怎么打出详细的错误log的。
--【伍】--： andy6688:
moderation
moderation参数传low试试，审核级别默认是auto，传low估计好一些。不过我一直传low的都有概率遇到你那个问题。gpt的审核一直都这么迷