API docs · Site 2 only
Every guide starts with a runnable quick start. Prices are read live from the model plaza; use “Open group” for full pricing, then create an API key for that group.
One row per model family. The same model can differ in price and features between groups; the group you pick decides.
One model set, two main channels: unsigned with custom sizes, or signed at 1K.
Text-to-image and reference edits with custom sizes; unusual ratios at 1K / 2K / 4K are upscaled. flare / sunburst are aliases here.
gpt-image-2gpt-image-2.5gpt-image-2.5-flaregpt-image-2.5-sunburstGroup #75 【GPT 网页端 逆向生图】image 2.0/2.5 主渠道
1K only with accurate ratios; images carry the OpenAI signature. Same API and parameters as the main channel with a key for this group.
gpt-image-2gpt-image-2.5gpt-image-2.5-flaregpt-image-2.5-sunburstGroup #159 【GPT 网页端 逆向生图】image 2.0/2.5 带签名
Native 4K, the Adobe channel, official OpenAI direct, or Banana rebuilding a GPT draft to 2K / 4K.
GPT Image 2.5 Flare / Sunburst at native 4K with multiple references and quality; see the guide and support matrix.
gpt-image-2.5-flaregpt-image-2.5-sunburstGroup #153 【GPT 生图】Image 2.5 全参 4K · 渠道 C
Also in #170
TestedBanana 2.1 chain 4K vs GPT Image 2.5 Flare / Sunburst 4KAdobe #127 support matrix and results
Adobe 4K channel for text-to-image and edits; quality (low / medium / high) must be set explicitly.
gpt-image-2Group #127 【GPT 生图】Image 2.0 2K (Medium)
Native 1K–4K straight from OpenAI's official API, billed per token, quality up to max. #185 goes through OAI and adds Flare / Sunburst; #184 goes through Azure with gpt-image-2 only.
gpt-image-2gpt-image-2.5-flaregpt-image-2.5-sunburstGroup #185 【GPT 官方直连 按量】OAI · Image 2 / Flare / Sunburst
Also #184 · Azure
GPT Image 2.5 drafts about 1K, then Banana 2.0 / 2.1 / Pro rebuilds it at 1K / 2K / 4K; per image, same price at every size.
gpt-image-2.5-banana2gpt-image-2.5-banana2.1gpt-image-2.5-banana-proGroup #177 【GPT 生图】Image 2 + 香蕉 2.1 High · 2K/4K（介意勿用）
TestedImage2 → Banana 2.1 rebuild at 2K / 4KBanana 2.1 chain 4K vs GPT Image 2.5 Flare / Sunburst 4K
Gemini native generateContent; click a card to open its guide. #139 returns image links by default and #125 returns Base64 by default. Models and prices are the same; a top-level response_format switches either way. Compare all request formats →
Text-to-image with single or multiple references; set ratio and 1K / 2K / 4K as in the samples.
gemini-3.1-flash-image-previewGroup #139 【原生生图】Nano Banana 主渠道 · 默认 URL
Generate images from text, images, PDFs or video as references; covers attachment formats, MIME and reading results.
gemini-nano-banana-2.1Group #139 【原生生图】Nano Banana 主渠道 · 默认 URL
Text-to-image and reference edits with its own model URL, parameters and save examples.
gemini-3-pro-image-previewGroup #139 【原生生图】Nano Banana 主渠道 · 默认 URL
Grok Imagine image edits.
Edit with 1–3 reference images; choose 1K / 2K, aspect ratio and response format.
grok-imagine-image-2.0Group #128 【Grok 生图】Grok Image 2.0
ByteDance Seedream 5.0 image model over the OpenAI Images endpoints.
Text-to-image, one image per request; billed per token, so larger sizes cost more.
seedream-5-proGroup #186 【Seedream 生图】Seedream 5.0 Pro · 按量
One video group #155, chosen by model; create a job, then poll it.
Text- or image-to-video; save the play / download links when the job finishes.
grok-imagine-video-1.5Group #155 【视频】多模型按秒计费
Finished jobs return play / download URLs valid for 1 hour, no key needed.
wan3.0-video-自部署Group #155 【视频】多模型按秒计费
Chat models over OpenAI, Anthropic and Gemini protocols.
Chat Completions, Responses, Anthropic Messages and Gemini native; paths are in the table at the end.
gpt-5.5grok-4.7gemini-3.1-prodeepseek-v4-proGroup #156 【文本模型】GPT 速刷/拥挤分组
See real-request quality, timing and parameter limits before you integrate. Reports open in a side panel.
Six real prompts: GPT 1K drafts against Banana 2.1 rebuilds at 2K and 4K and Lanczos upscales, with crops, PSNR and timings.
Three 4K options on the same prompts: full images and centre crops with real pixel sizes, timings and price per image.
generateContent parameter limits, attachment formats, MIME and item-by-item test results.
Measured limits for gpt-image-2 and 2.5 Flare / Sunburst: references, n, sizes, quality and output formats.
Supported ranges and edge results for references, n, resolution, quality, ratios and response formats, with real pixel sizes.
Errors users actually received in each group over the last 7 days. 401 / 403 mean a wrong, disabled or unbound key in every group.
| Status | Message | Cause and fix |
|---|---|---|
477 | 上游图片任务失败或需要审核，请更换提示词后重试 | The upstream rejected the job or it failed review. Change the prompt or reference and retry. |
400 | 非常抱歉，该提示可能违反了我们的内容政策…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
400 | 非常抱歉，该提示可能违反了关于裸露、色情或情色内容的防护限制…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
400 | 当前系统过载，如果高密集排队，请稍后重试 | The upstream queue is busy. Wait a few dozen seconds and lower concurrency. |
400 | 当前输入的图片地址生图服务无法访问…… | Reference URLs must be publicly reachable, not internal or expired signed links; or send a Data URL / multipart upload. |
503 | No available compatible accounts | Every account in the group is busy or cooling down. Retry shortly or use another group with the same model. #75 and #159 share the same models. |
| Status | Message | Cause and fix |
|---|---|---|
477 | 上游图片任务失败或需要审核，请更换提示词后重试 | The upstream rejected the job or it failed review. Change the prompt or reference and retry. |
400 | 非常抱歉，该提示可能违反了我们的内容政策…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
400 | 非常抱歉，该提示可能违反了关于裸露、色情或情色内容的防护限制…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
400 | 当前系统过载，如果高密集排队，请稍后重试 | The upstream queue is busy. Wait a few dozen seconds and lower concurrency. |
503 | No available compatible accounts | Every account in the group is busy or cooling down. Retry shortly or use another group with the same model. #75 and #159 share the same models. |
| Status | Message | Cause and fix |
|---|---|---|
400 | 本分组每次仅生成 1 张，n 只能为 1。 | Set n to 1 or omit it. Send parallel requests for more images. |
400 | 不支持该宽高比；支持 1:1、2:3、3:2…… | Use one of the 10 ratios listed in the message; the size tier follows the long edge. |
400 | 非常抱歉，生成的图片可能违反了我们的内容政策…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
502 / 503 | Upstream service temporarily unavailable · No available compatible accounts | Temporary upstream failure. Retry shortly; check usage records first so you do not generate twice. |
| Status | Message | Cause and fix |
|---|---|---|
503 | No available compatible accounts | Every account in the group is busy or cooling down. Retry shortly or use another group with the same model. #153, #158 and #170 share the same models. |
502 | Upstream service temporarily unavailable · Upstream request failed | Temporary upstream failure. Retry shortly; check usage records first so you do not generate twice. |
502 | Upstream access forbidden, please contact administrator | The upstream account is restricted. Contact the administrator; meanwhile use another group with the same model. |
451 | The generated images appear to be unsafe. Try modifying the prompts or the seeds. · prompt or image unsafe | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
429 | Upstream rate limit exceeded, please retry later | Upstream rate limit. Back off exponentially and lower concurrency. |
404 | Model "gpt-image-2" is not supported by any configured account in this group | The group does not offer that model (#158 has only 2.5 Flare / Sunburst). Use the names listed in the plaza. |
400 | This model is not supported on the Chat Completions endpoint | Call image models on /v1/images/generations or /v1/images/edits, not Chat Completions. |
| Status | Message | Cause and fix |
|---|---|---|
503 | No available compatible accounts | Every account in the group is busy or cooling down. Retry shortly or use another group with the same model. |
429 | Upstream rate limit exceeded, please retry later | Upstream rate limit. Back off exponentially and lower concurrency. |
400 | poll failed: … "image_unsafe" … The generated images appear to be unsafe…… | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
502 | Upstream access forbidden · Upstream service temporarily unavailable | Temporary upstream failure. Retry shortly; check usage records first so you do not generate twice. |
499 | (connection closed, no body) | The client disconnected before the image was ready. Adobe 4K often takes over a minute; set the client timeout to at least 300 s. |
| Status | Message | Cause and fix |
|---|---|---|
451 | prompt or image unsafe · The provided prompt is considered unsafe and it cannot be used to generate content. | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
502 | Upstream request failed after retries | Still failing after automatic retries. Try again later; 4K and multiple references take longer, so set the client timeout to at least 300 s. |
404 | Model "gemini-3.1-flash-image-preview" is not supported by any configured account in this group | The group does not offer that model name (common on #125). Use the names in the plaza, or switch to #139. |
400 | … INVALID_ARGUMENT … | The body does not match Gemini generateContent. Check contents, inlineData (plain Base64) and generationConfig against the guide. |
| Status | Message | Cause and fix |
|---|---|---|
400 | Generated image rejected by content moderation. | The prompt, reference or result was flagged by content moderation. Change the wording or reference and retry. |
451 | 内容审计命中风险规则，请调整输入后重试 | The input matched a risk rule. Change the prompt or reference and retry. |
400 | xAI upstream returned status … | The upstream rejected a parameter. Check: lowercase resolution 1k / 2k, no quality high, 1–3 references, a supported ratio (see the #128 matrix). |
| Status | Message | Cause and fix |
|---|---|---|
400 | task_not_exist | The job ID does not exist, or a different key was used. Poll with the same key that created the job. |
404 | Video request not found | The job is unknown or expired. Check the ID; regenerate once delivery URLs expire. |
502 / 503 | Upstream service temporarily unavailable · No eligible Grok media accounts | Video accounts are busy. Keep polling the job ID and create new jobs later. |
403 | This group does not allow /v1/messages dispatch | The video group only accepts the video endpoints, not Messages / Chat. |
429 | Upstream rate limit exceeded, please retry later | Upstream rate limit. Back off exponentially and lower concurrency. |
For chat or other clients, check the model and group in the plaza first; access depends on the key and the model.
| Protocol | Method and path |
|---|---|
| OpenAI Images | POST /v1/images/generations · POST /v1/images/edits |
| OpenAI Chat Completions | POST /v1/chat/completions |
| OpenAI Responses | POST /v1/responses |
| Anthropic Messages | POST /v1/messages |
| Gemini native | POST /v1beta/models/{model}:generateContent |
| Video jobs | See the Grok Video / Wan 3.0 guides |
Updated 2026-10-28 · Site 2 API docs
