I'm an under average skilled computer engineer and a novice fighting game player. I can speak Japanese a little bit. ex-@Hytopia, @StabilityAI, @Mozilla
Niigata Japan
Joined April 2008
- Tweets57.2K
- Following775
- Followers3.1K
- Likes27
Pinned Tweet
Request to your local publisher If you wanna read my book in English which explains UNIX V6 kernel in detail :D amazon.co.jp/dp/4774154644
Progress on the SNES game I'm having AI create. How should I put it? It feels like a poorly made RPG Maker game, and I just can't seem to get rid of that clunky feel.
AIにブラウザ上で動くSFC(SNES)のエミュレーターをRust+WebAssemblyで書かせ、さらにAIにその上で動くゲームを作らせてみている。
SNES Emulator: github.com/takahirox/snes-ru…
ゲームはまだ非公開
I've been experimenting with having AI autonomously iterate and improve toward a goal across around 10 projects, but iterative improvement really does consume a lot of AI quota0 usage.
思い出のパソコン通信PC-VANをブラウザ上で再現してみている。しかし如何せん記憶が曖昧なので実際はどんな感じだったか今ひとつ覚えていない。当時のユーザーだった方、もっとこんな感じだったとか教えてもらえると嬉しいです!
takahirox.github.io/pcvan/
I'm experimenting with defining a glTF extension to represent MMD in glTF.
github.com/takahirox/mmd-glt…
改めて色々調べてみていたんだけれど、僕の理解に誤りがなければやはり長期の住宅ローンってかなり美味しくないか?金利のある世界だからこそさらに手堅く美味しいという可能性すらある気がする。ローン全額払い終わったときに、家のお金を差し引いてもプラスになって終われることすらあると思う
Testing a browser extension to chain local LLMs via node graphs. Voice translation demo; other uses can balance speed and quality.
github.com/takahirox/my-audi…
I wanted to build a browser-based MSX emulator for years but never had the time. I asked AI to do it, and it finished in under an hour. I was really surprised, but emulator development is mostly about implementing specs, so it makes sense AI’s good at it.
github.com/takahirox/msx-rus…
Tryng to generate 3D model with Astra + Blender MCP + Reference Image(generated by Codex).
Not bad. Really not bad at all. Of course, it's not a perfect recreation, but the fact that a 3D model that's at least recognizable can be created as the intended character is a big deal.
The Rust and WebAssembly port of Three.js has come a long way, so as the next step, I think it's about time to start working on optimizations for Rust and WebAssembly.
takahirox.github.io/three-rs…
glTF lightmap extension proposal, yay
github.com/KhronosGroup/glTF…
Still hitting a wall with text-to-3D generation.
Trying to create 3D models using only Codex or Clotho subscriptions, without paying for extra services.
Simple voxel-style models work fairly well, but realistic humanoids are much harder.
Has anyone cracked this?
I have Codex generate a concept image, and use Codex or Claude, and Blender MCP to build a 3D model, then compare and iterate. Better images might help, but I'm starting to think the approach itself is flawed, especially for humanoid models. The results are far from what I want.






