Skip to content

gproxy v2.8.0

Choose a tag to compare

View all tags
@LeenHawk LeenHawk released this 16 Aug 01:12
v2.8.0
f8128c6

v2.8.0

Models Gemini Veo video generation and routes OpenAI video jobs to it; the OpenAI video family gains outbound endpoint synthesis.

English

Added

  • Veo wire models. Gemini :predictLongRunning requests (instances, parameters), the long-running operation poll (status, error, generated video references in both Gemini API and Vertex forms), and file download references are modeled in gproxy-protocol.
  • Video endpoint synthesis. request_target now synthesizes the OpenAI video family (create, retrieve, list, delete, download, remix, characters, edit, extend) and the Veo create/poll/file-download targets. Resource identifiers (video id, operation name, file id) travel in the model argument slot, like GetModel.
  • Video transforms. Create requests convert between OpenAI video jobs and Veo predict requests (size to aspect ratio/resolution, seconds to duration, data-URL references to inline image bytes); create and retrieve responses share a job-status mapping between OpenAI video objects and Veo operations. A completed operation surfaces its Veo file URI as gproxy_video_uri for download.

Fixed

  • Explicit unsupported pairs. Resolving a transform for web search, realtime, or an unrouted video operation now returns a structured unsupported-pair error instead of panicking.

简体中文

新增

  • Veo 协议建模。 gproxy-protocol 新增 Gemini :predictLongRunning 请求（instances、parameters）、长时操作轮询（状态、错误、Gemini API 与 Vertex 两种形态的生成视频引用）以及文件下载引用。
  • 视频端点合成。 request_target 现可合成 OpenAI 视频全家族（创建、查询、列表、删除、下载、混剪、角色、编辑、续写）与 Veo 的创建/轮询/文件下载端点。资源标识（视频 id、操作名、文件 id）沿用 GetModel 的 model 参数槽约定。
  • 视频转换。 创建请求在 OpenAI 视频任务与 Veo predict 请求之间转换（尺寸→宽高比/分辨率、时长映射、data URL 参考图→内联图像字节）；创建与查询的响应共享 OpenAI 视频对象与 Veo 操作之间的任务状态映射。完成的操作会把 Veo 文件 URI 以 gproxy_video_uri 透出供下载。

修复

  • 显式不支持配对。 对 web search、realtime 或未接线视频操作解析转换时,返回结构化的 unsupported-pair 错误而不是 panic。
Assets 59
Loading