🎧 Listen to this post as an audiobook
Audiobook in comic character voices (Qwen3-TTS, local)

A tidy image cockpit for Codex workflows just shipped. Pick a direction, front, three-quarter, side, three-quarter back, back, and it animates your character turning that way, now with experimental 16- and 20-frame takes. Handy. The catch hides in the word render: every frame is one image inference, which is to say one metered call. More angles, more frames, a smoother sprite, and a meter that spins right along with it. And the whole cockpit only flies on someone else’s engine.

Rendered Every Angle Except the Price

Source: RT @dreiachse: Image Cockpit for Codex Workflows v0.1.7 を公開しました! · twitter

What this means for ThakiCloud

The fun of the cockpit hides its real cost in that phrase, per-frame inference. Five directions at twenty frames is a hundred-plus metered calls to turn one character once, all of it printing on somebody else’s cloud. ThakiCloud pulls that math back inside your own walls, which is what on-prem means: the models run in racks you own. Metis serves the image and animation models on-prem, Paxis splits the directions and frames across worker agents, and however many frames you stack, what spins is your GPU, not a meter. Enjoy the cockpit. Just bolt the engine to your own rack.


An auto-generated comic riffing on this week’s industry news.

태그: animation, codex, image-generation, inference-cost, on-prem, sovereignty

카테고리:

업데이트: