工具概览
imagine-agy 是一个智能体技能和命令行工具,通过无头方式驱动 Google 的 Antigravity CLI(agy),为 Claude Code 增加图像生成能力。Claude Code 本身无法生成图像;该包通过 Antigravity 一等公民级的 generate_image 工具来路由图像生成,赋予它这一能力。它以 npm 包形式发布,既可作为 Claude Code 技能使用,也可作为独立的文生图 CLI 使用。
安装时运行 npx imagine-agy,它会把技能复制到 ~/.claude/skills/imagine-agy/,并在 ~/.claude/settings.json 中添加 Bash(agy:*) 允许规则。该 CLI 无需 Claude Code 也能使用,例如 npx imagine-agy "a red fox" -o fox.jpg。选项包括 --out、--ratio(1:1、16:9、9:16、4:3、3:4)、--model 和 --timeout。该包称每张图像的生成通常需要 30–60 秒。
该工具解决了无头 agy 图像生成中三个未被文档记载的失败模式:Claude Code 的 Bash 分类器拦截 agy、agy 在打印模式下自动拒绝自己的工具,以及 agy 忽略 shell 的工作目录。无论文件名如何,输出始终为 JPEG(基线、300 DPI),因此不支持透明背景。该技能增加了素材集批处理、长时间运行转后台、磁盘上验证以及单轴迭代。
定价
价格与限制可能变化,请以官网为准,并确认币种、付款周期、最低席位数和用量上限。
应用场景
- 团队用 imagine-agy 为落地页生成主视觉和横幅图。
- 团队用它制作博客和社交图片、插画、纹理、概念图、分镜和占位素材。
- 团队用它创建 9:16 比例的竖版手机壁纸。
- 团队用它在一次运行中批量生成整套素材,使主视觉、图标和背景在视觉上协调一致。
- 团队用它无需 Claude Code,直接在终端中生成图像。
- 团队用它在磁盘上验证生成的图像,而不是只相信运行摘要。
功能特性
根据文字提示生成图片
可作为 Claude Code 技能使用
提供文生图命令行工具
基于 Google Antigravity CLI
可制作主视觉图和图标
可在终端生成插画
用户评论
还没有评论,来分享你的使用体验吧
常见问题
Can Claude Code generate images?
Not on its own — Claude Code is a coding agent with no image model. This package gives it the ability by routing image generation through the Antigravity CLI, which does have one.
Does this need an API key?
No. It uses the existing Antigravity CLI sign-in, with no separate key to manage and nothing to configure.
Does it work outside Claude Code?
Yes. Running the CLI with a prompt and an output path is a standalone operation; the skill is optional.
Does it work with Cursor, Windsurf, or other AI editors?
The CLI works anywhere Node runs. The skill is in Claude Code's format, but any agent that can run shell commands can call the CLI directly.
How much does it cost?
It costs whatever the Antigravity plan charges for a generate_image call; the package adds nothing on top and has zero dependencies.
Can I generate transparent PNGs / logos with alpha?
No — output is JPEG, which has no alpha channel; the site points to its conversion options.
Can it put text in the image?
It can try, but the site advises against it because image models render words unreliably. Generate the art with space left for text and compose the type on top.
Can it edit an existing image or do img2img?
Not currently — the tool covers text-to-image only.
How many images can I generate at once?
The CLI does one per call. The skill batches a whole set in a single agy run, which also keeps them stylistically coherent.
Is this made by Google or Anthropic?
Neither. It is an independent open-source wrapper around the Antigravity CLI.