Agy image gen

Skill yinjs/agyenvoy/skills/agy-image-gen

Use when the user asks to generate or edit an image — illustration, photo, logo, icon, concept art — from a text description. Claude has no native image tool, but the agy (Antigravity/Gemini) MCP produces real raster images.From its SKILL.md

Install

npx -y skills add yinjs/agyenvoy --skill agy-image-gen

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.2 KB, as published. Nobody here has run it

Generating Images with agy

Overview

Claude has no image-generation tool. The agy (Antigravity) CLI wraps Google's Gemini native image model and can produce real raster images from a text prompt. This plugin exposes agy as the agy_ask MCP tool. When asked for an image, call agy_ask instead of refusing.

When to Use

User wants a generated image/illustration/photo/logo/icon/concept art from a description.
User wants a placeholder, hero, avatar, or asset image for a site or doc.
NOT for: charts/plots/diagrams from data (write code — matplotlib/mermaid/SVG), or flat geometric/vector shapes (agy writes code for those too — see Traps).

Recipe

Call agy_ask with the target directory in add_dir and an absolute save path in the prompt:

prompt: Using your native image generation model, create <rich photographic description>. Save the image to <absolute-dir>/<name>.png. Do NOT write a script or use matplotlib/PIL.
add_dir: ["<absolute-dir>"] ← same dir as the save path
timeout_ms: 600000 ← generation takes ~30–90s; give ≥5 min

Without both add_dir and an absolute save path, agy dumps the image to its own scratch dir instead of where you want it.

If agy_ask returns "not logged in", call agy_login first, then retry.

Traps

Symptom	Cause	Fix
Tiny PNG (few KB), flat shapes	Simple/geometric prompt → agy writes matplotlib/PIL code instead of calling the AI model	Say "Using your native image generation model" and describe a rich, photographic scene
Image saved elsewhere	No workspace set	Pass `add_dir` and an absolute path in the prompt
Command hangs / times out	Prompt referenced an input file not in the workspace	Only reference paths that exist; put any input image in the `add_dir` dir

Verify

Always confirm after the call:

file <name>.png   # expect: "PNG image data, 1024 x 1024, 8-bit/color RGB"

A real AI image is typically ~1 MB at 1024×1024. A few-KB file means the matplotlib trap fired — retry with the native-model phrasing.

Notes

Image generation selects the Gemini image model automatically — you don't pick it via the model param. model only affects text responses.
Editing an existing image uses the same recipe (reference the input path, which must be inside the add_dir dir). Verify the output actually changed.