Technical posts, build decisions, model experiments, honest mistakes. The Forge is open-source; the thinking behind it is here.
A running log of the decisions I'm making while building this studio. Drafts in progress, posts planned, nothing polished until it's published.
AnimeLegend God became AKASHI. FRX-1 became KAGE. A new third mascot, MIRA, joined. Naming is architecture. Here's the 23-file diff and what it taught me about naming as a discipline.
The pipeline is the product. After Effects is a glorious GUI; Remotion is a typed, testable, diff-able React component. Every shot is code. Here's why that matters when your studio is an agent team.
40 original style references. 3 hours of training. A LoRA that never saw a frame of copyrighted anime. The full prompt pack, the training config, the cost sheet, and the A/B against the base Flux.
Seven shots shipped, four needed a re-gen, one needed a whole new pipeline branch. I'm breaking down every failure — the prompt drift, the hand problem, the physics, the lip-sync. All of it, publicly, so the next short fails differently.
Same prompt, same seed, same style LoRA, two models. Where Pro wins, where dev still holds, and the one class of shot where I'm sticking with dev for now. Screenshots, cost per image, and the production decision.
Anime already thinks mythically. Startup marketing does not. The gap is an opening. Here's how I'm writing a studio that speaks like its own protagonist — and why the first draft felt like cringe until the fifth rewrite.
The log is ordered newest first. Planned posts are commitments — when a title sits here for more than four weeks without shipping, it means I changed my mind and will explain why in a follow-up.
Not ready for email? Catch the rest of the signal where it lands.
