genmedia 0.2.1

Multimodal media generation CLI for Google GenAI

genmedia

Multimodal media generation CLI for Google GenAI. Images, video, and image editing from the terminal. Bring your own API key.

Examples

Generate	Edit	Generate
genmedia image "a cat on a skateboard"	before editing	genmedia image "a glowing crystal orb floating in darkness"
genmedia image "a slightly glowing crystalline cat in the morning light in a forest, sharp, magical realism"	genmedia edit red_triangle.jpg "change to blue, yellow gradient background"

https://github.com/user-attachments/assets/46beecf4-a12a-4983-be7e-5ce516b8b803

Install

# uv (fastest)
uv tool install genmedia

# or run without installing
uvx genmedia image "a cat on a skateboard" --pretty

# pipx
pipx install genmedia

# pip
pip install genmedia

Requires Python 3.11+.

Setup

Get a Gemini API key and set it:

export GEMINI_API_KEY="your-key-here"

Usage

Generate images

# Simple
genmedia image "a cat on a skateboard" --pretty

# Full control
genmedia image "abstract gradient, soft pastels" \
  --model gemini-3.1-flash-image-preview \
  --size 4K \
  --aspect 16:9 \
  --output backgrounds/pastel.png

# Batch
genmedia image "product mockup on marble" --count 3 --output-dir shots/

# Using Imagen
genmedia image "photorealistic mountain lake" --model imagen-4.0-generate-001

Edit images

genmedia edit input.png "remove the background" --pretty
genmedia edit photo.jpg "make the sky more dramatic" -o edited.jpg

Generate video

# Text-to-video
genmedia video "a cat riding a skateboard downhill at sunset" --duration 8 --pretty

# Image-to-video (animate a still image)
genmedia video "the scene comes to life" --image keyframe.jpg --pretty

# Style reference (apply visual style from an image)
genmedia video "a bustling city street" --style-ref painting.jpg

# Asset reference (character/object consistency, up to 3 images)
genmedia video "the character walks through a forest" --asset-ref character.png

Note: --last-frame (frame interpolation) is available in the CLI but currently only supported by Veo 2 (veo-2.0-generate-001), which requires a Google Cloud Vertex AI billing account. It is not supported on Veo 3.x models.

List models

genmedia image --list-models
genmedia video --list-models

Piping

# Pipe prompt from stdin
echo "a neon cityscape at night" | genmedia image --pretty

# Write binary to stdout
genmedia image "a logo" --output - > logo.png

# Read image from stdin for video
cat keyframe.png | genmedia video "animate this" --image - --pretty

Dry run

genmedia image "test" --dry-run   # shows request payload without calling API

Output

JSON by default (for AI agents and scripts):

{
  "status": "success",
  "files": [
    {
      "path": "/tmp/genmedia/genmedia_001.jpg",
      "mime_type": "image/jpeg",
      "size_bytes": 534182
    }
  ],
  "model": "gemini-3.1-flash-image-preview",
  "elapsed_seconds": 14.3,
  "request": {
    "prompt": "a cat on a skateboard",
    "aspect_ratio": "16:9"
  }
}

Use --pretty for human-friendly output:

Saved to /tmp/genmedia/genmedia_001.jpg (521.7 KB)
Model: gemini-3.1-flash-image-preview | Time: 14.3s

Errors go to stderr as JSON with distinct exit codes:

• 0 — success
• 1 — API error (rate limit, server error, content blocked)
• 2 — validation error (bad params, missing API key)
• 3 — file I/O error

Models

Image generation

Model	Notes
gemini-3.1-flash-image-preview	Default. Best quality.
gemini-3-pro-image-preview	Previous generation.
gemini-2.5-flash-image	Older, faster.
imagen-4.0-generate-001	Imagen. Different API, good for photorealism.

Video generation

Model	Notes
veo-3.0-generate-001	Default. Standard quality.
veo-3.0-fast-generate-001	Faster, lower quality.
veo-3.1-generate-preview	Newer preview.
veo-3.1-fast-generate-preview	Newer fast preview.
veo-2.0-generate-001	Veo 2. Supports --last-frame. Requires Vertex AI billing.

For AI agents

genmedia is designed to be called by AI agents. JSON output by default, structured errors on stderr, distinct exit codes for branching. No interactive prompts, no spinners unless --pretty is set.

import subprocess, json

result = subprocess.run(
    ["genmedia", "image", "a logo for my app", "--aspect", "1:1"],
    capture_output=True, text=True
)

if result.returncode == 0:
    data = json.loads(result.stdout)
    image_path = data["files"][0]["path"]

License

MIT