GAMESLASHNew Effekseer VFX Skill Lets AI Agents Build and Review Godot Game Effects
By GameSlash
KanaWorks_AI has released an MIT-licensed VFX workflow for AI coding agents, with Effekseer compilation, a Godot capture lab, thirteen effects and separate visual review.
KanaWorks_AI published effekseer-vfx-skill on GitHub on October 10, 2026, with a further implementation update on October 11. The open-source project gives coding agents such as Claude Code and Codex a workflow to produce real-time game visual effects, compile them with Effekseer and inspect the result inside a Godot test scene. It is a game-development tool, not a new game or a feature built into Godot itself.
From a written brief to an actual rendered effect
The authoring skill moves through a design brief, texture work, a Python recipe, an EffekseerCore build, a Godot filmstrip and another iteration. The creator lists around thirty composable layer primitives and thirteen example effects across fire, ice, poison, dark projectiles, impacts, hit effects, slashes, charging, shields and healing. Each example is source material for developers to inspect; GameSlash has not run the builds.
Primary source: effekseer-vfx-skill repository
A separate reviewer checks visual results
A second skill, vfx-review, evaluates frame strips against a fixed rubric without seeing the author's recipe. Project examples include side-by-side comparisons showing how projectile smoke fades after the flame and how shield breakup can be made easier to read. Those comparisons illustrate the intended art-direction loop; they are not independent benchmarks.
What the project can and cannot do today
According to its README, authoring works on macOS, Windows and Linux with an appropriate Python environment. Compiling and rendering have been verified on Apple-silicon Macs and x64 Windows with Godot 4 and the Effekseer tooling. Linux rendering is not currently verified. The review measurements are image-based heuristics, not full GPU performance or overdraw measurements, and two of the thirteen examples were marked for another iteration.
The included case images show effects in a neutral test arena with two dummy targets. They are developer-produced before-and-after visuals, not captured gameplay from a shipped title. Example textures made using an image-generation model are disclosed in the project's third-party notices; the repository code is available under MIT, with third-party components retaining their own licenses.