genmedia 0.3.0

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.

Breaking changes in 0.3.0

• genmedia edit now takes the prompt as the only positional argument and one or more -i/--image flags: genmedia edit "merge them" -i a.png -i b.png (up to 14 images). The old genmedia edit input.png "prompt" form and the - stdin path are gone.
• --resolution 4K removed. The public Gemini API rejects 4K/2160p on every Veo model. Only 720p and 1080p are accepted, and 1080p requires --duration 8.
• --last-frame is now gated to veo-3.1-generate-preview only (the only model that actually supports it).
• veo-2.0-generate-001 removed from the model list (Vertex-only; genmedia has no Vertex backend).

New in 0.3.0

• genmedia video --negative-prompt "blurry, low quality" — things to avoid.
• genmedia image Imagen knobs: --guidance-scale, --person-generation (ALLOW_ADULT/DONT_ALLOW), --compression-quality (jpg only).
• genmedia image --size 1K/--size 2K now works on Imagen models (previously incorrectly rejected).
• genmedia edit accepts up to 14 input images for composition.

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 2K \
  --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 "remove the background" -i input.png --pretty
genmedia edit "make the sky more dramatic" -i photo.jpg -o edited.jpg

# Compose multiple images (up to 14)
genmedia edit "place the character in the forest" -i character.png -i forest.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 only supported by veo-3.1-generate-preview. --resolution 1080p requires --duration 8. Use --negative-prompt "blurry, low quality" to steer Veo away from artifacts.

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

# (Note: `genmedia edit` no longer accepts stdin — pass files with `-i`.)

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. Multi-image edit/composition (up to 14).
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. Only model that supports --guidance-scale, --person-generation, and --compression-quality (jpg only).

Knob compatibility:

Flag	Applies to
--guidance-scale	Imagen models only
--person-generation	Imagen models only (ALLOW_ADULT, DONT_ALLOW; ALLOW_ALL is Vertex-only)
--compression-quality	Imagen models only, requires --format jpg
--size	All image models (Imagen + Gemini) — 512, 1K, 2K
-i (multi)	gemini-3.1-flash-image-preview for genmedia edit composition

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. The only model genmedia allows for --last-frame (frame interpolation).
veo-3.1-fast-generate-preview	Newer fast preview. --last-frame is currently disabled here pending API confirmation; file an issue if you need it.

Knob compatibility:

Flag	Applies to
--negative-prompt	All Veo models
--last-frame	veo-3.1-generate-preview only
--resolution 1080p	All Veo models, requires --duration 8
--resolution 720p	All Veo models
--style-ref, --asset-ref	All Veo models

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