Asset Gen is the asset half of Godogen. It tells an agent where each kind of asset should come from: GPT-Image-2.6 for images, sprites and animated sprites, local generators for 3D, rigging, animation, sound and voice, Blender for exact shapes, and paid APIs for music and special voices. Your stated budget decides which route it takes.
Scripts handle image generation, grid slicing of sprite kits, loop-frame detection and seamless audio loops. Every generated asset is tracked in the project’s README with its real size in the scene, which is where a 3D model’s scale is decided.
What You Get
- Guidance on which source fits each asset type: images, transparent sprites, GLB models, rigged humanoids, sound effects, voice lines and music
- A Python script, asset_gen.py, for image generation and the paid APIs, plus helpers for grid slicing, audio looping and loop-frame search
- A character pipeline from transparent reference image through textured GLB and Mixamo rig to idle, walk, run and custom moves
- A Blender recipe for exact or articulated 3D built by a headless script
- An asset manifest table in the project README with size, path and cost for each asset
Common Use Cases
- Produce a consistent set of 2D sprites and backgrounds for a game
- Turn a reference image into a rigged, animated 3D character with lip-synced speech
- Generate sound effects, ambience loops and voice lines for a game or a video
- Estimate and confirm the cost of a paid image set before generating it
- Build a mechanical 3D model with moving parts in Blender
What Makes It Unique
It makes spending a decision the user controls: local tools are free, paid calls are priced and confirmed first, and the brief’s words about budget steer the route. It also records each asset’s size in the scene because generated models come out one unit tall. A style-family rule asks the agent to make one hero asset first and derive the rest from it.
Under The Hood
SKILL.md routes the agent to api.md for the paid services and motion.md for gameplay-driving animation, and to Python scripts it runs from the project. It reaches the network to call image, 3D, speech and music services from OpenAI, xAI, Google and Tripo with API keys that you supply as environment variables, and to download packages. For local generation the agent may set up a separate godogen_assets project that needs an NVIDIA GPU, a large amount of disk and a Hugging Face token, and runs GPU tools one at a time. It installs Python packages and Blender, runs Blender headless scripts, and writes generated files and the README manifest into your project. Paid calls cost real money and the skill tells the agent to confirm before the first one.