AGENT SKILL

create-image-gemini

暂无简介

使用场景图像与多媒体
License 许可
Apache-2.0
所属仓库
markhayden/bakin
最近更新
2026-11-13
Stars
★ 3
Forks
0

03 SKILL.md 正文

Create Image With Gemini

Use this skill when the image should be generated through a Google Gemini image route. The images plugin prefers configured runtime routes first, then falls back to the direct Gemini adapter when a direct API key is configured.

Inputs

  • taskId: required task id for asset linkage
  • brief: creative brief
  • surface: optional surface profile
  • model: optional Gemini image model, defaulting through routing. Prefer gemini-3.1-flash-image-preview for general work and gemini-3-pro-image-preview for premium routes when configured.
  • quality: optional quality tier

Procedure

  1. Build a provider-neutral prompt packet with concrete subject, style, lighting, camera/composition, and visual constraints.
  2. Call bakin_exec_images_recommend with provider: "google" plus the surface, model, objective, and quality.
  3. Confirm the returned route uses provider google.
  4. Call bakin_exec_images_generate with provider: "google", the returned model, surface, quality, task id, prompt, and prompt packet.
  5. Return assetId, provider, model, surface, width, height, and promptHash.

Retry Discipline

  • If the generate call times out or the transport result is ambiguous, do not change the prompt, route, surface, model, quality, or task id.
  • Before retrying, call bakin_exec_assets_list with the same taskId and type: "images" and reuse any matching generated asset for that task.
  • If no matching asset exists, retry bakin_exec_images_generate once with the exact same parameters. Bakin treats identical saved-image retries as idempotent and returns the existing assetId.

Do not call Gemini directly. The images plugin owns runtime/native routing, authentication lookup, asset save, and generation metadata.

04 同仓库更多 Skills