bug: 图片生成失败， WebSocket error: rate_limit_exceeded - Image rate limit exceeded

[jesssecat]

问题描述

使用最新代码无法生成图片

复现步骤

我看其他的issue已经解决了，但是我使用最新的代码，用docker-compose方式和代码运行方式都无法生成图片，使用的图片模型有：grok-imagine-1.0、grok-imagine-1.0-fast

预期行为

可以正常生成图片

实际行为

无法正常生成图片

部署方式

本地 uv run

版本/提交号

16e37b1

相关日志/报错

2026-03-31 10:47:32 | INFO     | app_chat.py:50 - AppChatReverse proxy is empty, requests will use direct network
2026-03-31 10:47:33 | INFO     | chat.py:292 - Chat connected: model=grok-3, stream=True
2026-03-31 10:47:37 | WARNING  | image.py:199 - App-chat image collect failed, falling back to ws_imagine: %s
2026-03-31 10:47:37 | INFO     | ws_imagine.py:105 - Image generation: prompt='一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微...', n=1, ratio=1:1, nsfw=True
2026-03-31 10:47:38 | INFO     | ws_imagine.py:220 - WebSocket request sent: 一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微笑...
2026-03-31 10:47:42 | WARNING  | ws_imagine.py:283 - WebSocket error: rate_limit_exceeded - Image rate limit exceeded
2026-03-31 10:47:42 | WARNING  | image.py:444 - WS batch failed: Image rate limit exceeded
2026-03-31 10:47:42 | WARNING  | image.py:463 - Image finals insufficient (0/1), running 5 recovery attempts for remaining=1, parallel_enabled=True
2026-03-31 10:47:42 | INFO     | image.py:484 - Recovery using 2 distinct tokens
2026-03-31 10:47:42 | INFO     | ws_imagine.py:105 - Image generation: prompt='一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微...', n=1, ratio=1:1, nsfw=True
2026-03-31 10:47:42 | INFO     | ws_imagine.py:105 - Image generation: prompt='一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微...', n=1, ratio=1:1, nsfw=True
2026-03-31 10:47:44 | INFO     | ws_imagine.py:220 - WebSocket request sent: 一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微笑...
2026-03-31 10:47:44 | INFO     | ws_imagine.py:220 - WebSocket request sent: 一位中年学者站在古色古香的书房里，周围环绕着高高的木质书架，上面摆满了线装古籍和卷轴。他面带温和的微笑...
2026-03-31 10:47:48 | WARNING  | ws_imagine.py:283 - WebSocket error: rate_limit_exceeded - Image rate limit exceeded
2026-03-31 10:47:49 | WARNING  | ws_imagine.py:283 - WebSocket error: rate_limit_exceeded - Image rate limit exceeded
2026-03-31 10:47:49 | WARNING  | image.py:504 - WS recovery batch failed: Image rate limit exceeded
2026-03-31 10:47:49 | WARNING  | image.py:504 - WS recovery batch failed: Image rate limit exceeded
2026-03-31 10:47:49 | INFO     | image.py:514 - Image recovery attempts completed: finals=0/1, attempts=5
2026-03-31 10:47:49 | ERROR    | image.py:520 - Image generation failed after recovery attempts: finals=0/1, blocked_parallel_attempts=5
2026-03-31 10:47:49 | WARNING  | exceptions.py:123 - AppException: server_error - Image generation blocked or no valid final image
2026-03-31 10:47:49 | ERROR    | response_middleware.py:75 - Response: POST /v1/images/generations - 502 (17150.22ms)
2026-03-31 10:53:15 | WARNING  | exceptions.py:123 - AppException: invalid_request_error - The model `grok-imagine-1.0` is required for image generation.
2026-03-31 10:53:15 | WARNING  | response_middleware.py:75 - Response: POST /v1/images/generations - 400 (2.73ms)
2026-03-31 12:42:22 | INFO     | scheduler.py:50 - Scheduler: starting token refresh...
2026-03-31 12:42:22 | INFO     | scheduler.py:54 - Scheduler: refresh completed - checked=0, refreshed=0, recovered=0, expired=0

请求样例（可选）

[shengzengyang]

用玩法模式生成的图片是不是模糊的那种,然后显示502 模型速率限制好像是,就是限流

[shengzengyang]

我让我的Claude 去帮我看了下,他认为就是上游的x.ai那边做出的限制,没有办法处理,询问一下,你的grok图片生成是否正常

…

On Wed, Apr 1, 2026 at 11:07 AM michaelbrinkworth ***@***.***> wrote: *michaelbrinkworth* left a comment (chenyme/grok2api#416) <#416 (comment)> From this stack trace, this is a rate limit issue with grok. Running npx ai-doctor could help. It can fix weak retry strategy with exponential backoff, jitter, retry caps, and retry-after handling. — Reply to this email directly, view it on GitHub <#416?email_source=notifications&email_token=BKDUUUUYWFE6WSATIMQ72DT4TSBX5A5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWG4YTCNRQGU4KM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4167116058>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BKDUUUTQLJEMT7UABK3U5LL4TSBX5AVCNFSM6AAAAACXHF343KVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCNRXGEYTMMBVHA> . You are receiving this because you commented.Message ID: ***@***.***>