image-gen

SkillMedia

Generate a single image from a text prompt using the MiniMax image generation API. Use when the user asks to create, generate, or render an image from a prompt and explicit width/height.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the image-gen skill

What this skill tells your AI

The instructions your AI receives, as published by sma1lboy/rove in .claude/skills/image-gen/SKILL.md and read by ahel’s review.

Generate one image from a text prompt via MiniMax image-01 model. The output file is written to the filename you pass as the 4th argument (relative to the caller's cwd).

The script path is resolved relative to this SKILL.md (not the caller's cwd), so it works from any directory. It accepts exactly four positional arguments: prompt, width, height, output_filename.

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
python3 "$SCRIPT_DIR/scripts/image-gen.py" "<prompt>" <width> <height> <output_filename>
ArgumentTypeConstraints
promptstring≤ 1500 chars; describe subject + style + mood
widthint512–2048, must be a multiple of 8
heightint512–2048, must be a multiple of 8
output_filenamestringFilename to write the image to, relative to cwd (e.g. img1.png). The file extension is preserved as-is.

Output

  • File: <output_filename> (relative to cwd — the directory the command is invoked from)
  • Format: raw image bytes base64-decoded from the API response; extension is whatever you pass (e.g. .png, .jpeg)
  • Always overwrites an existing file at the same path

After running, confirm the file exists with ls -lh <output_filename> and report the absolute path to the user.

Prompt writing tips

Be explicit about style to avoid photorealistic defaults. Example style keywords:

  • Flat / UI / vector: UI style, flat illustration, vector art, minimalist, clean cut-and-paste aesthetic, no photography no realistic textures
  • Collage / cut-and-paste: paper cut collage, layered flat cut-out shapes, torn paper edges, mixed media
  • Vintage / retro: vintage paper collage, sun-faded tones, retro travel poster
  • Photorealistic: photorealistic, documentary photography, film grain

Combine subject + style + palette. Example structure: "<style keywords>: <subject>, <key visual elements>, <color palette>, <negative constraints>"

Examples

1. Santorini paper-cut collage (UI style, non-photorealistic)

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
python3 "$SCRIPT_DIR/scripts/image-gen.py" "UI style paper cut collage of Santorini, Greece: whitewashed houses with blue domes, ancient Greek columns, olive branches, terracotta rooftops, layered flat cut-out shapes with torn paper edges, clean vector illustration, minimalist cut-and-paste aesthetic, pastel color palette, no photography no realistic textures" 512 512 santorini.png

2. Square UI icon (1:1)

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
python3 "$SCRIPT_DIR/scripts/image-gen.py" "flat vector icon of a paper airplane, minimal geometric shapes, soft blue and white palette, clean UI illustration, no shadows no gradients" 512 512 airplane.png

3. Landscape banner (16:9)

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
python3 "$SCRIPT_DIR/scripts/image-gen.py" "minimalist flat illustration of mountains at sunset, layered geometric shapes, warm orange and purple palette, clean vector style, no texture no realism" 1280 720 mountains.png

4. Portrait poster (2:3)

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
python3 "$SCRIPT_DIR/scripts/image-gen.py" "vintage paper collage portrait of a woman in 1920s fashion, torn paper layers, muted sepia and teal tones, art deco cut-out shapes, mixed media aesthetic" 832 1248 portrait.png

Signals

GitHub stars
122
Forks
8
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
image-gen-sma1lboy
Source
github.com/sma1lboy/rove