Asset tools¶

GodotMaker generates art via small Python helper scripts. /gm-asset calls them automatically — but you can also run them by hand if you need to regenerate a specific file or experiment with different settings.

asset_gen.py¶

asset_gen.py is the main API-backed image and 3D model generator. It supports Gemini, OpenAI, xAI Grok, and Tripo3D. Runtime-native native / codex image generation is selected by /gm-asset, not this script.

Image providers:

Provider	Cost per image	Best for
xAI Grok	2 cents (1K or 2K only)	Fast, good for most sprites and UI elements
Google Gemini	5–15 cents (512 to 4K)	More precise prompt following, better for detailed or edited images
OpenAI	5–7 cents in the local budget model	OpenAI Images API generation/editing

Provider-prefixed selectors require the matching key: GOOGLE_API_KEY / GEMINI_API_KEY for Gemini, OPENAI_API_KEY for OpenAI, and XAI_API_KEY for Grok. To choose which provider /gm-asset uses, set asset_image_model in ../06-configuration/project-config.md.

Generating an image¶

python tools/asset_gen.py image \
  --prompt "top-down pixel art player character, blue outfit, 64x64, transparent background" \
  -o assets/sprites/player.png

To use Grok instead:

python tools/asset_gen.py image \
  --prompt "top-down pixel art player character, blue outfit, 64x64, transparent background" \
  --model grok \
  --size 1K \
  -o assets/sprites/player.png

To edit an existing image (image-to-image):

python tools/asset_gen.py image \
  --prompt "add a glowing aura around the character" \
  --image assets/sprites/player.png \
  -o assets/sprites/player_glow.png

Common options:

Option	Default	Notes
`--prompt`	(required)	Describe what you want
`--model`	project config	`gemini[:model]`, `openai[:model]`, or `grok[:model]`; overrides `asset_image_model`
`--size`	`1K`	OpenAI: `1K`. Grok: `1K`, `2K`. Gemini: `512`, `1K`, `2K`, `4K`
`--aspect-ratio`	`1:1`	Many options — run `--help` to see all
`--image`	none	Provide a reference image to edit
`-o`	(required)	Output file path

Setting a budget¶

You can cap total spending to avoid surprises:

python tools/asset_gen.py set_budget 500

This sets a 500-cent ($5.00) limit, tracked in assets/budget.json. Any generation command that would exceed the remaining budget exits with an error before making an API call.

Generating 3D models¶

The glb subcommand converts a PNG image into a 3D model (.glb file) using Tripo3D. This requires TRIPO3D_API_KEY and is only relevant for 3D games.

python tools/asset_gen.py glb \
  --image assets/sprites/tree.png \
  -o assets/models/tree.glb

Cost is roughly 40–50 cents per model depending on the --quality preset (default or high).

rembg_matting.py¶

rembg_matting.py removes solid-color backgrounds from images, producing PNG files with transparent backgrounds. Use this when you have a sprite rendered on a plain background and need it cut out for use in a scene.

# Single image, auto-detect the best approach
python tools/rembg_matting.py assets/sprites/enemy_raw.png -o assets/sprites/enemy.png

# Batch: process all PNGs in a folder
python tools/rembg_matting.py --batch raw_frames/ -o clean_frames/

# Generate a preview so you can check the result before using it
python tools/rembg_matting.py assets/sprites/enemy_raw.png --preview

The tool uses a neural network (BiRefNet) to identify the subject and color matting to clean up the edges. It auto-selects the right approach for most images; you can force a mode with -m trust, -m adapt, or -m color if the automatic result is not clean enough.

GPU acceleration is used automatically if an NVIDIA GPU with CUDA is available. On CPU it is slower but still works.

tripo3d.py¶

tripo3d.py is the Tripo3D API client used by asset_gen.py glb internally. You do not normally call it directly — use the glb subcommand of asset_gen.py instead. It requires TRIPO3D_API_KEY.

Calling these by hand¶

You usually do not need to run these scripts directly. /gm-asset orchestrates them based on your ASSETS.md and calls them with the right prompts and output paths automatically.

The main reasons to call them manually:

You want to regenerate one specific asset with a tweaked prompt.
You are experimenting with different sizes, providers, or aspect ratios before committing to a full /gm-asset run.
You received art from an external source and want to remove its background with rembg_matting.py.

If you want to update the visual targets that /gm-evaluate checks against, re-run /gm-asset rather than editing the generated images directly. Editing them by hand will be overwritten the next time /gm-asset runs.