You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[feat] Integrate the OpenAI Images API (gpt-image) #139
Integrate the OpenAI Images API into OpenDirector, offering gpt-image and other image models confirmed available in the documentation as an optional image generation service in the existing creation workflow.
Scope
Verify the endpoint, authentication, model identifiers, request parameters and response format against the official API documentation, and cite the documentation in the implementation notes.
Use the existing provider and model configuration mechanisms to let users configure an OpenAI API key and select a model. Keep this service optional and do not change the default provider.
Support text-to-image generation. Expose editing, variants and other capabilities through the existing capability mechanism only after official interface support is confirmed.
Integrate generated results with the existing canvas and asset management workflows.
Handle authentication failures, rate limits, timeouts and server errors; do not leak API keys in logs.
Document configuration, usage, supported capabilities and limitations.
Acceptance criteria (implementer/reviewer to perform and provide evidence)
With valid credentials, image generation can be initiated from OpenDirector and the result displayed, saved and reused.
Invalid credentials, rate limits and timeouts produce clear feedback without affecting other providers.
Provide test records and end-to-end screenshots or video, identifying the actual model and API version used.
This issue records a product requirement; implementation, acceptance and evidence are the responsibility of a future assignee. Related: #138
Activity
kunkunblueberry commentedon Nov 6, 2026
接取本 issue,正在实现中(分支 eat/openai-images-gpt-image)。
计划交付范围:
esponse_format(该参数会被服务拒绝),兼容服务(vLLM-Omni Z-Image-Turbo 等)行为不变。
ate_limit_exceeded → 限流、�illing_hard_limit_reached → 额度、moderation_blocked → 内容政策,日志不记录 Key。
与 #140 是同一需求的英文版,一次实现同时覆盖两者。真实 API Key 下的端到端出图与截图证据仍需要凭证,会在 PR 中明确标注为未覆盖项。PR 链接随后补在这里。
kunkunblueberry commentedon Nov 6, 2026
实现已提交:#155
包含 openai-gpt-image profile(不发送
esponse_format、参考图重复 image[]、16 张 / 25 MB / 50 MB 上限、三个文档尺寸、�ackground: transparent 抠图)、invalid_api_key/
ate_limit_exceeded/�illing_hard_limit_reached/moderation_blocked 分类、docs/guides/openai-images.md 与 13 项 MockTransport 测试。
未完成项:真实 API Key 下的端到端出图与截图证据,需要凭证,PR 中已明确标注为未覆盖。验证结果:
pm test 483 passed、pytest 673 passed(53 项本机既有环境失败,已用 stash 基线对照确认与本次改动无关)、契约检查通过。
[feat] provider: add OpenAI Images (gpt-image) as an optional image s…
[docs] provider: record the live gpt-image run against a compatible r…
kunkunblueberry commentedon Nov 6, 2026
已用真实 Key 在 OpenAI Images 兼容中转站上完成端到端验证(2026-11-06),细节也写进了 PR #155 与 \docs/guides/openai-images.md\。
实际调用:模型 \gpt-image-2.5-1k\(该站上 4 个模型均为 \gpt-image*\,你给的 \oai-image\ 不在模型列表里),接口 /v1/images/generations\ 与 /v1/images/edits\,连接类型 API。
结论:\openai-gpt-image\ profile 的请求形状(不带
esponse_format\)被真实服务接受,多图的 \image[]\ 提交也被接受,鉴权失败分类正确且不泄漏 Key。
两点限制:① 该中转站忽略请求的 \size\,按自身默认尺寸出图,OpenDirector 以实际探测尺寸记录素材;② OpenAI 官方端点本身未验证,计费/限流表现需官方 Key 复核。证据来自 headless 运行,为日志与素材尺寸,没有界面截图。
[-][feat] 对接 OpenAI Images API(gpt-image)[/-][+][feat] Integrate the OpenAI Images API (gpt-image)[/+][feat] provider: add OpenAI Images (gpt-image) as an optional image s…
[docs] provider: record the live gpt-image run against a compatible r…
[feat] provider: OpenAI Images (gpt-image) provider and a label taxon…